Weather Shocks, Coping Strategies and
Consumption Dynamics in Rural Ethiopia
Jianfeng Gao and Bradford F. Mills
This Version: November 2016
Abstract:Household panel data is used to estimate the impact of weather shocks on consumption in rural Ethiopia, along with the effectiveness of household coping strategies in ameliorating the impact of shocks. Results show that changes in rainfall levels are positively associated with per adult equivalent consumption,while high temperature is associated with lower consumption. In terms of household coping strategies, public transfers mitigate the impact of against adverse rainfall shocks on consumption, and off-farm employment mitigate the impact of high temperature shocks. However, urban migration and transfers from former household members or informal social safety nets do not mitigate adverse weather shock.
Keywords:resilience, climatic shocks, consumption, Ethiopia
JEL Codes: Q54, Q12, D13
1
1 Introduction
Adverse (poor or variable) weather conditions have been shown to reduce the mean yields of agricultural productsand increase the output variance in developing countries (Cabas, Weersink, and Olale 2010; Felkner, Tazhibayeva, and Townsend 2009; Fisher et al. 2012; Kaylen, Wade, and Frank 1992; Schlenker et al. 2009; Schlenker and Roberts 2006; Thornton et al. 2009). When households rely heavily on rain-fed agriculture, induced production shocks often translate into income shocks and, in turn, into negative consumption shocks. Rural households adopt a wide range of strategies to mitigate the negative impacts of adverse climatic shocks. Common resiliency strategies includeprecautionary savings to smooth consumption (Paxson 1992) and diversification into income generating activities less vulnerable to climatic shocks, including migration (Marchiori, Maystadt, and Schumacher 2012; Barrios, Bertinelli, and Strobl 2006), off-farm employment (Ito and Kurosaki 2009; Bezabih et al. 2010), and cropping varieties resistant to heat and drought (Phiri and Saka 2008). Ex-post, households may sell livestock or productive assets during hard times (Dercon 2002; Zimmerman and Carter 2003; Kazianga and Udry 2006). Asset sales often lower future earnings potential and, thus, are seen as a negative coping strategy (Del Ninno, Coll-Black, and Fallavier 2016). Households also make use of formal or informal social safety nets to mitigate consumption impacts of negative shocks(Pan 2009; Fafchamps 1992; Fafchamps 2011).
Previous studies have examined the effectiveness of individual coping strategies such as precautionary savings(Paxson 1992), migration(Taylor, Rozelle, and de Brauw 2003; de Brauw and Harigaya 2007), off-farm employment (Kochar 1999), crop diversification (Gatiso 2015), asset sales(Fafchamps, Udry, and Czukas 1998; Kazianga and Udry 2006), and formal and informal social safety nets (Pan 2009; Quisumbing and McNiven 2010), but to the best of our knowledge no study to date has presented a comprehensive evaluation of different coping strategies in terms of their effectiveness in mitigating the impact of adverse climatic shocks. Further, the literature shows thatindividual coping strategies employed by households often do not fully mitigate the adverse impacts of weather shocks on household welfare. For example, Dercon(2004) finds persistent negative impacts of rainfall shocks on per capita consumption in rural Ethiopia.This leaves the crucial question of which coping strategies successfully buffer against adverse weather shocks and which strategies do not.A systematically evaluation of coping strategies can identify successful existing strategies, and assistpolicy makers and development agencies in devising aid programs and interventions that help rural households better cope with climate change.
The objectives of this paper are to assess the impact of weather shocks on household consumption in rural Ethiopia, and to evaluate the effectiveness of widely used coping strategies in mitigating weather shock impacts. This study differs from previous studies in several important aspects. First, it systematically evaluates the effectiveness of a basket of rural household coping strategies in buffering against weather shocks. We show that several coping strategies employed by rural Ethiopian households are effective, but in combination they only partially mitigate the impact of adverse weather shocks on consumption. Second, we construct a new dataset and employ novel empirical strategies to generate more reliable estimates of the weather impacts on household consumption, whichfind patterns different from previous studies such as Porter (2012). Third, our results are used to simulate weather shock and coping strategy impacts on household poverty, and to suggest that more effective programs and policies to assist rural households in Ethiopia in increasing their resilience to external weather shocks.
Ethiopia is particularly interesting for the study. The country’s economy is dominated by its agriculture sector, which accounts for 43% of GDP and 90% of exports.[1] Agriculture is primarily rainfed and thus highly dependent on rainfall, whichaccording to USAID (2015), is increasingly erratic, with marked seasonal deficits and more frequent drought and heavy rainfall events. In the past four decades alone, devastating droughts occurred in 1973-74, 1983-84, 1987-88, 1990-91, 1993-94 and 2015-16.[2]On the other hand, households in Ethiopia employ a variety of strategies to cope with weather shocks, including participating in the Productive Safety Net Program (PSNP) - one of the strongest formal safety nets programs in Sub-Saharan Africa (SSA). These variable weather conditions and widely used coping strategies allow us to identify the impacts of weather shocks on household consumption and the effectiveness of household coping strategies to mitigate the adverse weather impacts.
The remainder of this paper is structured as follows: section 2 describes the data and the Ethiopian context; section 3 outlines the conceptual and empirical framework; section 4 presents the main results and associated robustness tests; and section 5 concludes the paper.
2 Data and Context
Household-level data from the Ethiopian Rural Household Surveys (ERHS)[3]are joined with village-level rainfall data to form a unique panel dataset, which contains detailed information on consumption, income, and idiosyncratic shocks of approximately 1,500 households in 15 rural villages (kebeles, wards, or peasant associations (PAs)) from 1994 to 2009, as well as local historical daily rainfall records from 1980 to 2009.
2.1 Household data
The households were surveyed twice in 1994, and subsequently in 1995, 1997, 1999, 2004 and 2009, giving a sample of about 1500 households in 15 villages across the country (see Figure 1 for locations). The villages were selected to account for diversity in the farming systems in Ethiopia, and within each village households were sampled through a stratified random sample. We use household-level panel data from the 1994, 1999, 2004 and 2009 rounds to form an equally spaced panel dataset, with a total sample size of 5,673 observations.
The 15 villages covered in the ERHS are characterized by seasonal and fluctuating rainfall(Gray and Mueller 2012). Rainfall occurs mainly during the Kiremt season in the summer, and in some villages also during a secondBelg season in the spring. Average annual precipitation in these villages ranges from 470 to 1300 mm (18 to 51 inches). Historically, severe droughts occurred in 1999, 2002–2003, 2005, and 2008, and their adverse effects could not be fully mitigated by the government’s social protection programs.
The structured questionnaire administered to each sample householdcollected information on household demographics, assets, expenditures, agricultural activities, as well as community level data on electricity and water, sewage and toilet facilities, health services, education, NGO activity, migration, wages, and production and marketing. Many questions remain the same across survey panel rounds.
The survey has two notable features: (1) Attrition at the household level is very low at 1.3% per year and 13.2% in total between 1994 and 2004. (2) Although only 15 of the thousands of villages in rural Ethiopia were sampled, they are broadly representative of households in non-pastoralist farming systems, and many of the average health and nutrition outcome variables in ERHS were very similar to those in the nationally representative Welfare Monitoring System collected by the Central Statistical Office in 1994 (Dercon and Hoddinott 2011).
2.2 Weather data
Climatic data were drawn from the African Flood and Drought Monitor (AFDM)[4], which contains precipitation (mm), maximum temperature (K), and minimum temperature (K) on a daily basis and with a grid resolution of 0.25 decimal degrees. The rainfall data are at the village level as approximated by the inverse distance weighting interpolation method, using weather data from the four nearest grids around the village. Thus, rainfall is treated as a covariate village shock to village households.
The main climatic variables used in this study are: (1) the average daily rainfall in the main rainy season (June 16th to September 15th) in the year prior to the survey[5], (2) the “standard deviation” of daily rainfall in the main rainy season in the five years ending in the year before the survey,[6] (3) total growing degree days (GDD) for the whole growing season (April 1st to September 30th) in the year prior to the survey, and (4) total extreme heat degree days (EHDD) in the whole growing season in the year prior to the survey. Daily rainfall is first averaged for the main rainy season level in each year, and these yearly rainfall data are used to calculate for each panel period the “standard deviation” of average daily rainfall in the main rainy season over the past five years. GDD measures accumulated exposure to an optimal growing temperature range over the growing season, thus is positively correlated with crop production. By contrast, EHDD measures accumulated exposure to extreme heat above the upper threshold of the optimal growing temperature range, thus is often found to be negatively correlated with crop production, offering a good indicator for an adverse temperature shock (see Schlenker and Roberts 2006; Roberts, Schlenker, and Eyer 2013; Schlenker et al. 2009 for more discussion). Total GDD and EHDD in the whole growing season are derived using daily maximum and minimum temperatures, as described in Appendix A in Gao and Mills (2016).
2.3 Variables and summary statistics
The dependent variable is the logarithm of real monthly consumption per adult equivalent for each household.[7] Following Porter (2012) and Dercon, Hoddinott, and Woldehanna (2012), the monthly consumption measure consists of food consumption (including food expenditure and value of food received as gifts) and non-investment non-food consumption (excluding investment type consumptions such as durables, health and education expenditure). Some food consumption data are projected from consumption over a one-week recall period to make the aggregate household consumption comparable across survey rounds. The monthly nominal consumption measure is then deflated by a food price index (FPI) constructed from village level data collected at the same time as the household survey (see Dercon and Krishnan (1998) for details). The adult equivalence scales are based on nutrition (calorie) needs for different age and gender groups as guided by the World Health Organization (WHO). A detailed scale table can also be found inDercon and Krishnan (1998).
Table 1 presents the summary statistics for real monthly consumption per adult equivalent from 1994 to 2009 in 1994 prices. Mean real consumption in 1994 is 85.80 birr per adult equivalent per month, and steadily rises to 105.34 birr in 1999 and to 116.00 birr in 2004. In 2009, mean real consumption sharply drops to 70.52 birr per adult equivalent per month, due to severe droughts that occurred in several villages in Tigray region and Southern Nations, Nationalities, and Peoples' Region, and dramatic world staple food price increases in 2008 (Dercon, Hoddinott, and Woldehanna 2012).
Tables 2 reports the summary statistics of the independent variables included in the analyses. On average, a household has six members, with a real monthly consumption per adult equivalent of 96 Birr and a livestock holding of three livestock units at one year prior to the survey. A large share, 71%, of the households have a male head, and 10%, 15%, and 9% of the households experience illness, input price, and death shocks between survey rounds, respectively.
Figure 2 presents the rainfall patterns between 1980 and 2009 across villages. As shown, rainfall in the study villages varies considerably across years. In terms of average levels, rainfall in northern villages (such as Haresaw and Geblen)is substantially lowerand more variable than in other villagessuch as Yetmen and Imdibir (Figure 3). Daily minimum and maximum temperatures fluctuate little over time, but show considerably heterogeneity across space (Figure 4). Total growing degree days and total extreme heat degree days show greater variation over time than temperatures, with total extreme heat degree days appearing to generally increase and become more variable in the early 2000s, but there is also significant spatial heterogeneity in the patterns(Figure 5).
Data on the average adoption rate of common household coping strategies is presented by year from 1994 to 2009 in Table 3. Overall, engaging in off-farm activities is the most prevalent (33% of the households), followed by receiving public transfers (21%), receiving transfers from ISSN (17%), and sending migrants for labor market reasons (15%). Receiving remittances has the lowest adoption rate at 3%. All coping strategies has increased between 1994 and 2009, with income from transfers from informal social safety nets, family remittances, and off-farm activities increasing steadily over the years. A large proportion of households that have sent migrants to urban areas did not receive remittances: for instance, in 2009 20.42% percent of households sent migrants to urban areas for economic reasons, but only 2.1% reported receiving remittances. The adoption of coping strategies varies over time following institutional change. The PSNP, for instance, was launched in January 2005 as the successor to the Employment Generation Scheme, leading to an increase in percentage of households reportedly receiving public transfers. Import bans imposed by the Arab States of the Persian Gulf in 1998 and 2000, along with regulations prohibiting the sale of land, proscribing the loss of land rights for migrants and imposing registration requirements for new migrants, may limit the interregional movement of labor and cause the percentage of households sending migrants much lower in 1994 and 1999 than in 2004 (Dorosh 2013). Both information of the strategy mix employed and the increasingprevalence of strategies employed over time in our sample make rural Ethiopia a particularly interesting case study of the effectiveness of alternative coping strategies.
3 Conceptual and Empirical Frameworks
The benchmark models of household consumption dynamics are the permanent income model and the full insurance model(Morduch 1995). Both models imply that income shocks will show a low correlation with changes in household consumption if households have access to insurance, credit, or liquid assets and if income shocks are predominantly transitory in nature (Bardhan and Udry 1999). When these conditions are not met, the household will employ the coping strategies like asset sales, use of informal and formal assistance networks, and activity diversification to mitigate the impact of shocks to the extent possible. In the case of weather shocks, agricultural production is directly affected, causing an income shock. Observing or anticipating this shock, the household adopts coping strategies to minimize adverse impacts. If coping strategies are fully effective, then income will be stabilized and consumption will not be affected. Otherwise, the income shock will be transformed into a household consumption shock. Figure 6 outlines these links between pre-shock economic wellbeing (consumption), shock exposure, coping strategies, and post-shock economic wellbeing.
This study focuses on five common coping strategies rural households in Ethiopia adopt: (1) sending migrants to urban areas; (2) applying for and receiving transfers from the government (public transfers); (3) requesting and receiving transfers from an ISSN; (4) receiving remittances from former household members; and (5) engaging in off-farm activities. These strategies have not been previously examined as a comprehensive portfolio of household options.
3.1 Benchmark empirical models
Low rainfall and high temperature impacts on household consumption are identified with a fixed-effect panel data model. The mitigating impacts of coping strategies onclimatic shock estimates is then examined.
The impact of weather shocks on consumptionis first assessed in the following empirical specification:
(1)
where
per adult equivalent consumption of household in village at time ;
permanent income of household in village at time ;
a vector of characteristics of household in village at time ;
a vector of weather conditions in village at time ;
a vector of idiosyncratic shocks to household in village at time ;
time indicator;
household fixed effects;
idiosyncratic error term.
The weather indicatorsconsidered in the specifications include: average daily rainfall if it is above the historical mean and if it is below the historical mean; the standard deviation of rainfall if it is above the historical mean and if it is below the historical mean; two dichotomous variables indicating whether average daily rainfall is below the historical mean and whether the standard deviation of rainfall is above the historical mean; the logarithm of total growing degree days in the whole growing season; and the logarithm of total extreme heat degree days in the whole growing season. Livestock assets one year prior to the survey, ,are used as a proxy for permanent income(Dercon, Hoddinott, and Woldehanna 2012; Porter 2012). As an important asset of rural households, livestock not only determine the productive capacity of the households, but also signal the wealth of the households, and thus provides a good indicator of permanent income potential.Time-varying household characteristics include household size and composition and demographic indicators of the household head.Idiosyncratic shocks that may affect household income include (1) illness, if any household member was sick in the past five years; (2) death, if any household member died in the past five years; and (3) input prices, if there was large increase in agricultural input prices in the past five years.