Booms, Busts, and Household Enterprise: Evidence from Coffee Farmers in Tanzania

Abstract

Smallholder agricultural commodity suppliers in developing countries are often vulnerable to global commodity price fluctuations. Using panel data on farmers from an area of Tanzania where most farmers grow coffee, this study finds that global coffee prices matter for household outcomes, through their effects on farmgate prices, coffee sales and revenues, and household expenditures. The article documents that households cope with coffee price busts by increasing enterprise ownership, an effect that is greater for households without access to other means of coping. Comparisons of mean outcomes of enterprises operated by coper households (which operated an enterprise only in periods of low coffee price) with those of stayer households (which operated an enterprise throughout the sample period) indicate that the former are less likely to be profitable or to hire workers.

1. Introduction

As price-takers in global commodity markets, smallholder farm households are often vulnerable to unpredictable events a world away (Blouin and Macchiavello 2013). This is especially true in countries where government protections against price volatility are weak (Van Hilten, Fisher, and Wheeler 2011). Household vulnerability is exacerbated by the fact that although markets for savings, credit, and insurance exist in low-income agricultural contexts, they often function very poorly (Burgess and Pande 2005; Cole et al. 2013; Dupas and Robinson 2013; Karlan et al. 2014). Informal coping mechanisms, such as intra-village or intra-household transfers (Townsend 1994) and labor-related migration (Bryan, Chowdhury, and Mobarak 2014; Morten 2016; Dinkelman 2017), also exist but are often imperfect. Furthermore, if a large proportion of households in a village produce the same agricultural commodity, price shocks of that commodity can create aggregate macroeconomic shocks that are difficult to insure against within the village.

How do smallholder commodity producers cope with global price fluctuations? This article reports on a study of this question in the context of rural households in Tanzania responding to global coffee prices. In the sample, over 80 percent of households reported harvesting coffee at least once, indicating the relative importance of this commodity to the local economy. The study examines outcomes for entire communities in the sample, including households that are not growing coffee.

By linking detailed panel survey data on households to a time series of coffee prices, global prices are found to robustly predict farmgate prices, quantity sold, and revenues from coffee (for households that farm coffee), as well as household food and non-food expenditures (for the entire community). The results show that to cope with price busts, households resort to small-scale enterprise activity. A drop of one standard deviation in the global coffee price increases the probability of enterprise ownership by about 5 percentage points, or 13 percent above mean ownership. Most of this response takes the form of low-investment businesses (e.g., roadside vendors selling farm goods) and is concentrated among households with fewer physical and financial assets.

The finding that a significant fraction of households engaged in enterprise activity are unlikely to expand into large businesses is consistent with previous studies.¹McCaig and Pavcnik (2016) used national representative panel data from Vietnam to document the dynamic characteristics of small-scale enterprises and the relative importance of owner characteristics. They found that most microenterprises do not transition into formal or high-growth enterprises (a finding that is also consistent with the work of Schoar [2010]), although initial success predicts later ability to hire workers and transition into formality.² All these stylized facts are consistent with this study’s result that enterprise is used as a coping mechanism for some households in agricultural contexts.

This article documents two facts in support of the hypothesis that enterprise is also an imperfect coping mechanism. First, households with greater physical and financial assets, which can be sold for cash when farm profits are low, are significantly less likely to open a business during coffee price busts. The average effect is driven by households without access to these other (potentially more effective) mechanisms. Second, enterprises used for weathering shocks (i.e., businesses that are open only during coffee price busts) perform poorly compared to enterprises that operate more consistently (i.e., throughout the time period of the panel). These intermittent business owners use less labor and working capital and realize lower profits.

This study makes three contributions. First, it adds to the understanding of microenterprises in developing countries. McCaig and Pavcnik (2016), as mentioned previously, showed that most microenterprises do not transition into high-growth or formal enterprises, though early success and owner’s human capital predict transition into formality and increased employment. Here it is shown that household enterprise can function as a coping mechanism for households whose primary income is not from entrepreneurship, though it seems to be used mostly by less wealthy households.

Second, this article adds to the literature on coping mechanisms in low-income contexts. Previous studies have demonstrated that households undertake a variety of measures to mitigate the deleterious effects of income shocks. Using the same dataset as in the present article, Beegle, Dehejia, and Gatti (2006) showed that households cope with income shocks (crop loss) by increasing child labor. Other means of coping, from different contexts, include savings (Paxson 1992), wage labor (Kochar 1999; Dimowa, Michaelowa, and Weber 2010), and temporary migration (Bryan, Chowdhury, and Mobarak 2014; Dinkelman 2017; Morten 2016). This study adds household enterprise to the set of coping mechanisms. It also finds that households in the sample may be relying on some of the other smoothing mechanisms highlighted in previous studies in addition to enterprise ownership, although those results are more imprecisely estimated, possibly due to measurement error in the timing of the income shock variable. The results of this study are not unique to agricultural household coping behaviors in response to global commodity price shocks, but are likely generalizable to other income and productivity shocks faced by such households. Indeed, Adhvaryu and Nyshadham (2017) have documented similar behaviors in response to temporary, acute health shocks to household members.³

Third, the evidence presented in this article complements the macro literature on non-agricultural self-employment. One key stylized fact in this literature is that self-employment and transitions from unemployment to self-employment are strongly countercyclical (Bosch and Maloney 2010; Fiess, Fugazza, and Maloney 2010; Loayza and Rigolini 2011; Koellinger and Roy Thurik 2012; Shapiro 2014). Here this is shown to be true for self-employment in agricultural areas as well.

The remainder of the article is organized as follows. Section 2 describes the world market for coffee and provides institutional details on coffee production in Tanzania. Section 3 describes the dataset used in this article and the construction of primary outcomes and regressors. Section 4 presents the empirical strategy and discusses its validity. Section 5 reports results from the empirical tests of the main predictions of the model. Finally, section 6 concludes.

2. Context

Coffee is a major commodity traded in the global market, with trading worth US$16.5 billion occurring in 2010. About 90 percent of coffee is produced in developing countries (Ponte 2002), the majority of it by smallholder farmers (Brown, Charveriat, and Eagleton 2001). Coffee is one of Tanzania’s largest exports, and the country produces about 0.8 percent of world output (Board 2012).⁴

A main concern for identification is that households react to coffee prices by changing the intensity of coffee farming, or by starting or stopping coffee farming altogether, based on the global price. Since coffee trees generally take more than three years to produce their first fruit, short-term entry and exit into coffee farming are not usually possible (though intensive margin decisions such as whether to harvest or how to allocate labor might be important margins of adjustment). It is verified that in the data used in this study, global coffee price fluctuations did not change selection into the coffee grower sample or affect the acreage under coffee.

Cooperative unions and primary societies at the village level have traditionally been the main institutions undertaking procurement of coffee from farmers (Baffes 2005). The Tanzania Coffee Marketing Board is the primary government body in charge of regulation of the coffee industry. Prior to the 1990s (before the study period), farmers received advance payments on delivery based on a preannounced price by the Board. After sales at the auction, the Board would deduct fees and transfer the remaining revenues to the cooperative unions, who then deducted their input credits if any and their processing costs before transferring the remainder to primary societies. The primary societies deducted their own costs, and if there was money left over, this was given to the farmers. In the early 1990s (the study period), the government introduced some reforms in the sector, allowing unions to decide the advance and the total payment. Econometrically, using year and month fixed effects ensures that the estimates in this article are not affected by policy changes during the study period. Section 4 discusses the estimation process in detail.

3. Data

This study uses survey data from the Kagera region of Tanzania, an area west of Lake Victoria and bordering Rwanda, Burundi, and Uganda. Kagera is mostly rural and primarily engaged in producing bananas and coffee in the north and rain-fed annual crops (maize, sorghum, and cotton) in the south. The Kagera Health and Development Survey (KHDS) was conducted by the World Bank and Muhimbili University College of Health Sciences. The sample consists of 816 households from 51 clusters (or communities) located in 49 villages covering all five districts of Kagera. Each household was interviewed up to four times between the fall of 1991 and January 1994 at six- to seven-month intervals. The randomized sampling frame was based on the 1988 Tanzanian census.

It is important to note that while each household was sampled every six to seven months, surveying occurred essentially continuously in the study areas as teams of enumerators cycled through households. This process enables coffee-related activities to be captured at small time intervals and allows the full variation in global coffee prices over time to be used across the household panel.

A two-stage, randomized stratified sampling procedure was employed. In the first stage, census clusters (or communities) were stratified based on agro-climactic zone and mortality rates and were then randomly sampled. In the second stage, households within the clusters were stratified into “high-risk” and “low-risk” groups based on illness and death of household members in the 12 months before enumeration and were then randomly sampled. There was moderate attrition from the longitudinal sample: 9.6 percent of households sampled in wave 1 were lost by wave 4.⁵ However, to preserve balancing across health profiles in the sample, lost households were replaced with randomly selected households from a sample of predetermined replacement households stratified by sickness. KHDS is a socioeconomic survey following the model of previous World Bank Living Standards Measurement Surveys.

The survey covers individual-level, household-level, and cluster-level data related to the economic livelihoods and health of individuals, as well as the characteristics of households and communities. About 80 percent of the households in the entire sample reported harvesting coffee at least once in the survey period (1991–93), indicating the importance of coffee in the local economy.⁶ The analysis includes all surveyed households, since non-coffee farmers would also be plausibly affected by coffee price fluctuations. The supplementary online appendix presents regression results of coffee prices faced by coffee-growing households on international prices, showing that international prices affect local prices. Results for the main outcomes for only the coffee-growing sample are also reported in the supplementary online appendix.⁷ The KHDS data are combined with data on monthly international coffee prices available with the International Coffee Association.

The monthly prices are prices of robusta coffee, which is the main variety of coffee grown in the Kagera region. The following subsections outline the variables used in the analysis. A more detailed definition of each of the variables can be found in supplementary online appendix. Tables 1 and 2 present summary statistics for international coffee prices as well as the household-level variables used in the analysis.

Price Lag Variable

The first wave of the survey asked households about their economic and labor activities in the 12 months preceding the survey. The second, third, and fourth waves, however, asked households about their economic and labor activities in the last six months. This is because the time lag between waves was about six to seven months, so the questions were changed to avoid asking about overlapping time periods. In order to estimate the impact of international coffee price fluctuations on a household, the outcome variables were matched to the appropriate international price faced by the household at the time when it was making decisions regarding labor allocations and microenterprise ownership.

As information is available on the month and year in which households were surveyed, the average international price for the time period about which the survey asked could be used. In the first wave, this was the average price over the last 12 months preceding the survey month of the households, and for the subsequent waves it was the average price over the last six months. For example, if a household was interviewed in wave 1 in September 1991, the price faced by the household is the average international robusta coffee price from September 1990 to August 1991. If a household was interviewed in any wave other than the first, the price faced by the household is the average international robusta coffee price over the preceding six months; for example, if it was interviewed in September 1993, prices from March 1993 to August 1993 would be considered. A drawback of the lag structure of the price variable is that it smooths out very short-run price fluctuations and makes it more difficult to identify very transitory coping mechanisms.⁸

The average price computed in this manner is about 46 cents/lb, with a standard deviation of about 3.9 cents/lb.⁹ The independent variable of interest is the lagged robusta price divided by its standard deviation over the survey period. The coefficient on this variable is the marginal effect of a one-standard-deviation change in the price.¹⁰

An alternative definition of the price variable would utilize information on location-specific coffee-harvesting seasons and only use price lags for the previous season in a particular location. While this would capture the price variation affecting farmers more precisely, the KHDS only asks about the outcomes of interest for the last 12 months in the first wave and the last six months in the remaining three waves; therefore a season of two to three months would not be picked up precisely in these measures.

Household-Level Variables

At the household level, this study examines the impact of coffee prices on revenues from coffee, consumption expenditure, and microenterprise ownership. Because surveys were carried out after about six months following the first survey, the period to which the survey questions pertain is the last 12 months for the first wave and the last six months for the subsequent waves. The area harvested for coffee is on average only about 10 percent of the total area harvested by households, but annual revenues from coffee sales account for approximately 43 percent of agricultural revenues for the sample, and the proportion increases to 67 percent if only households reporting nonzero coffee revenues are included. Thus, coffee sales make up a significant component of household income.

Regarding microenterprise ownership, almost 40 percent of the households reported owning an enterprise at some point over the four waves of the survey. As table 2 indicates, about 42 percent of households reported never owning an enterprise, and about 12 percent owned an enterprise in all four waves. Over half of the enterprises are undertaking trading or other informal non-farm self-employment, referred to here as low-investment businesses. High-investment enterprises are those that require skilled or semi-skilled labor, and include enterprises such as stallkeeping and restaurant ownership to professions such as blacksmith, plumber, or carpenter. For a full description of the included categories, refer to supplementary online appendix. The main distinction between these two types of enterprise is that low-investment businesses require relatively little or no investment in fixed or human capital.

Summary statistics are presented separately for four categories of households. The first category consists of households that owned at least one enterprise in all four waves of the survey (not necessarily the same enterprise). While these might not be the same enterprise, their ownership indicates that a household is more likely to rely on entrepreneurship as a consistent source of income rather than as a coping mechanism. Such households are referred to as “stayer” households. About 12 percent of the households in the sample (123 households) are stayer households. Households that owned an enterprise at least once in the survey period but are not stayers are labeled “switcher” households, because they switched enterprise status during the course of the panel. Switcher households make up nearly 50 percent of the households in the sample (447 households). The switchers are further divided into two subcategories. The first comprises households that owned an enterprise only when the coffee price was low, defined as when the price was below its 25th percentile value for the survey period; these are labeled “coper” households, as it is posited that they use enterprise as a means of weathering income shocks. As indicated in table 2, about 6 percent of households (54 households) are copers. The second subcategory includes all other switcher households, which make up about 44 percent of the sample (393 households).¹¹ The intensive margin outcomes for these categories of households will be examined to test whether the households make different enterprise decisions and have differently performing enterprises, as well as to study the relative success of using enterprise as a coping mechanism. Table 1 presents summary statistics on ownership, intensive margin outcomes, and household and financial characteristics of the whole sample as well as the four categories of households. As will be discussed in the following paragraphs, the greatest contrast among intensive margin variables exists between the stayers and the copers, with most of the intensive margin variables for the other switcher households lying in between. Note that as these variables are considered conditional on owning an enterprise, their values are not driven mechanically by the fact that stayer households owned a business for longer periods.

The intensive margin variables studied comprise three categories of enterprise operations: capital assets, labor, and performance. The first category is composed of three variables: a binary variable for whether the enterprise owns a capital asset, a binary variable for whether the enterprise bought or sold a capital asset, and the total input expenditure in the survey period. The majority of enterprises, about 73 percent, own a capital asset, although the proportion is larger for the stayers, about 87 percent, and relatively low for the copers, about 43 percent.¹² Average input expenditures are around 3,093.87 Tanzanian shillings (TZS) for the whole sample. The expenditures of copers are 501.553 TZS, and those of the stayer households are over 10 times greater, at around 5,578.45 TZS.

The labor category comprises three variables. The first is the number of weeks spent in self-employment during the survey period for all household members who reported working in self-employment in the last seven days, aggregated up to the household level. The second labor category variable is a binary variable for whether a household member helped in the enterprise, which on average was true for about 36 percent of the entire sample, and ranged from 28 percent for the copers to nearly 42 percent for the stayers. The third variable is a binary variable for whether a hired worker was employed in the enterprise.

The performance category consists of a binary variable for whether the business had positive profits in the reference period.¹³ On average about 55 percent of enterprise-owning households reported positive profits for at least one of their enterprises.

Table 1 also lists summary statistics for the characteristics of the head of household and measures of the household’s financial resources. These characteristics are taken from the first wave in which each household appears and are treated as fixed for the purposes of exploring heterogeneous enterprise activity responses to coffee price fluctuations in table 5. Head of household characteristics include gender, literacy and numeracy skills, and a binary variable that equals 1 if the household head underwent some primary schooling and equals 0 otherwise. Financial resource measures comprise six measures, which indicate whether a household sent and received remittances, whether it had positive savings, whether it had positive debt, and whether the household’s financial and physical stocks were above the sample median values of financial and physical stocks, respectively. As table 2 indicates, stayer households are more likely to have a head of household who is literate and numerate relative to the whole sample, especially the coper households. Their measures of financial resources are also higher relative to the whole sample; for instance, they have a 0.7 probability of owning financial stock greater than the median value in the sample. The gap is large relative to the coper households, which have a 0.47 probability of owning financial stock greater than the median value in the sample. Table 2 presents the ownership histories wave by wave, in conjunction with coffee prices.

4. Empirical Strategy

The empirical analysis proceeds in three stages. It begins with determining (i) the extent to which global coffee prices matter for the sample of coffee farmers, and (ii) how price shocks impact expenditures, enterprise ownership, and performance for all households in the sample. The first question is investigated by checking whether global prices are correlated with the farmgate prices that coffee farmers face, and whether the global prices affect quantities of coffee harvested, coffee revenues, and household expenditures on food and non-food items.¹⁴

As described in the previous section, the price p varies at the month|$\, \times \,$|year level. Households surveyed in the same month of a particular wave will therefore face the same (retrospective) coffee price; households that happen to have been surveyed in different months of the same wave will face differing prices. When coffee prices, coffee revenues, or area under coffee are the dependent variables, the sample is restricted to coffee-growing households only. For all other outcomes, the entire sample of surveyed households is used. Standard errors in all regressions are clustered to allow arbitrary correlation in the error term at the level of the enumeration cluster (which is the primary sampling unit).

To estimate how fluctuations in the coffee price affect business ownership, an ownership dummy is regressed on the coffee price as well as the month, year, and household fixed effects using the model specified in equation (1). A distinction is also made between low-investment and high-investment businesses, as defined in the previous section. These two ownership variables are regressed in separate specifications to study whether sensitivity of business ownership to coffee price fluctuations is different across business types.

Third, equation (1) is estimated for other possible means of weathering shocks, such as savings, debt/loans, and remittances. As mentioned in section 3, the fact that only responses over six or seven months to coffee price averages over the same period can be estimated implies that short-run coping mechanisms cannot be precisely identified. Therefore, these mechanisms could form important transitory coping mechanisms but be imprecisely estimated.

5. Results

This section presents the results of the empirical analysis described in the previous section. The aim is to understand household responses in enterprise activity to fluctuations in agricultural profitability deriving from the global price of coffee.

Effects of Global Prices on Farming Decisions and Expenditures

The regression analysis starts by showing that global coffee prices affect outcomes for the coffee farmers in the sample.¹⁵ The results are reported in table 3. First, the relationship between the global price and the farmgate price, imputed from the transactions data, is investigated.¹⁶ The results, reported in column (1) of table 3, demonstrate that the farmgate price is sensitive to movements in the global price: an increase of one standard deviation (SD) in the global price increases the farmgate price by about 0.08 SD.¹⁷

Next is an examination of the effects of coffee price fluctuations on whether a household sold a positive quantity of coffee and whether it received positive coffee revenues. The results, reported in columns (2) and (3) of table 3, show that a one-SD change in the global price of coffee increases the probability that a household sold some coffee and received some amount of coffee revenues by about 8 and 5.6 percentage points, respectively. The estimates are significant at the 1 and 5 percent levels, respectively. To remove influential outliers, both farmgate prices and expenditures are trimmed at the 1st and 99th percentiles; results with untrimmed expenditures are presented in table S1.4 in the supplementary online appendix.

Lastly, the extent to which households are able to smooth consumption in the face of coffee price fluctuations is explored. Total household expenditure (for all households, not only coffee farmers) is regressed on the global coffee price, and evidence is found against perfect smoothing, consistent with the previous literature. Total expenditures increase by about 9,600 TZS (from a mean of roughly 187,000 TZS) among coffee-farming households for every one-SD rise in the coffee price (table 3, column 4). Expenditures are then split into food and non-food categories, and for both categories changes in expenditures corresponding to movements in the coffee price are found. Both food expenditures (column 5, approximately 3,000 TZS from a mean of about 36,000 TZS) and non-food expenditures (column 6, approximately 2,200 TZS from a mean of about 64,000 TZS) reflect variations in response to shocks in the global coffee price. While both food and non-food expenditures are impacted by coffee prices, the effects as a proportion of the mean coffee price are much larger for food expenditures, perhaps reflecting short-term liquidity constraints. Results for the expenditure and enterprise ownership outcomes for the sample of coffee-growing households only (i.e., households that reported harvesting coffee at least once over the sample period) are reported in table S1.2 in the supplementary online appendix.

Despite the large impacts of coffee price shocks on household revenues and expenditure, price variations do not significantly predict movements along either the extensive (table 3, column (7)) or the intensive (table 3, column (8)) margins of coffee growing. That is, column (7) shows that a one-SD increase in the price of coffee insignificantly reduces the probability of a household harvesting some amount of coffee by 0.7 percentage points, while column (8) shows that the same price increase drives households to harvest 0.02 acres more of coffee; neither is statistically significant. These results are consistent with the fact that since coffee trees take two to three years to start producing coffee, short-term quantity adjustments or extensive margin selection into coffee growing are not possible. Thus, the effects on revenue found in previous paragraphs are driven by farmgate price changes, not quantity adjustments.

Both the above results can be interpreted as evidence of no short-run relationship between coffee price and coffee growing.¹⁸ To mitigate any possible selection not covered by these robustness checks, the analysis is focused on main outcomes, such as participation in the enterprise sector and labor and capital allocations across sectors, for all households.

Household Enterprise Activity and Coffee Price Fluctuations

Having verified the impacts of coffee price shocks on household revenues and expenditures, the next step is to examine whether global coffee price changes affect the probability of non-agricultural business ownership among coffee-growing households. The results of this analysis are reported in table 4. Column (1) shows results of a regression of an enterprise ownership dummy on the coffee price. It is found that a one-SD rise in the global price decreases the probability of enterprise ownership by nearly 5 percentage points, or around 12 percent of mean ownership. This finding can be interpreted as evidence of countercyclical household entrepreneurship in the sample. On average, households are more likely to engage in enterprise activity during coffee price busts, and to shut their businesses during coffee price booms.

Columns (2) and (3) examine ownership of high- and low-investment businesses. The results show that households are more likely to cope with income variations due to coffee price shocks using low-investment businesses. A one-SD rise in the coffee price leads to a drop of 3.8 percentage points in the probability of a household owning a low-investment enterprise. Ownership of high-investment businesses, on the other hand, does not vary significantly with the coffee price, and the coefficient is about one-third the size of the impact of coffee price on low-investment business ownership.

Next, the degree to which this enterprise activity response to coffee price fluctuations among coffee-growing households varies by household financial resources is explored. That is, to the degree that intermittent enterprise activity appears to be an income shock mitigation mechanism for some households in the sample, one might suspect that this response would be most pronounced among households constrained in other obvious dimensions of mitigation (e.g., buffer financial stock, divestible physical capital, and access to debt). Accordingly, table 5 reports the heterogeneous effects of coffee price fluctuations on enterprise activity by various dimensions of financial resources. Coffee price is interacted with physical asset stock, financial savings, loans issued, debt received, and total asset stock. Total asset stock is equal to the sum of physical asset stock, savings, and loans issued minus debt received. In each specification reported in table 5, the main effects of the financial variables and the coffee price are included in addition to their interactions.

It is found that the enterprise response to coffee price fluctuations indeed varies by financial resources of the household. Households with greater resources (higher physical and financial asset stocks, less debt) are less likely to increase their enterprise activity in response to coffee booms. This heterogeneity in the effects of coffee price on enterprise is significant across all five measures of household resources. The effect could be because households with greater resources are less affected by coffee price shock (for instance, due to their occupational choices or economic diversification), or because they have more resources to cope with the shock ex post. Perhaps of interest is the observation that the estimates of the main effects of these resource measures indicate that households with greater resources are less likely on average to own enterprises; however, it is unclear which way the causation runs (i.e., it cannot be determined if households with greater assets choose not to start enterprises or if households with enterprises are either likely to draw down their assets or less likely to accumulate assets in the first place). Accordingly, the main effects of resources are not interpreted here, but rather only their interactions with the exogenous coffee price fluctuation.

Other Means of Coping

The final investigation is of the degree to which coffee price fluctuations drive households to use other sectoral participation or production decisions as coping mechanisms. As mentioned in section 3, the fact that six-month lags (for waves 2–4) and 12-month lags (for wave 1) of the coffee price are used as the main regressor implies that it is harder to pick out short-term coping mechanisms. However, table 6 shows that while most coefficients are not statistically significant, the directions of the coefficients are as expected: during periods of higher coffee prices, households are more likely to save and have lower debt levels and are also less likely to receive remittances or engage in outside employment (other than agriculture on their own land or self-employment); furthermore, they are less likely to harvest other crops, an effect that is statistically significant. The effect on household size, on the other hand, is small relative to the mean household size and is not statistically significant. Taken together, the results seem to suggest that intermittent enterprise activity is an important means of weathering income shocks deriving from coffee price fluctuations in this setting, while some other mechanisms, such as increasing net savings, are also important.

Additional Robustness Checks

A primary concern with the analysis is that attrition from the sample, which may be correlated with the coffee price, could affect the results. Table S1.1 in the supplementary online appendix presents results for the main outcomes—household expenditures and enterprise ownership—for those households that reported at least some data in all four waves. These results are consistent with the main results, indicating that attrition is not driving the results.¹⁹

Finally, to test whether the expenditure results are driven by local inflation caused by coffee price increases and not reflected in real expenditures, two additional tests are conducted. First, household expenditures controlling for cluster-by-year fixed effects (as well as for month and household fixed effects as before) are regressed on the coffee price; any cluster-by-year inflation effects should be absorbed by these fixed effects. The results, reported in table S1.3 in the supplementary online appendix, are robust to the inclusion of the additional fixed effects. Second, table S1.3 also reports household expenditures after adjustment for inflation using two price indices calculated in the KHDS. Columns (4)–(6) use the price index based on all predicted prices for the missing prices, while columns (7)–(9) use the price index based only on actual prices for those that were not missing (Ainsworth et al. [2004] detail the construction of the price indices). The correlation between the two indices is approximately 0.99, and the results are robust to this adjustment.

6. Conclusion

This article’s analysis of coffee production, coffee revenues, and household expenditures indicates that households in this context are poorly insulated from shocks to the global coffee market. Since smallholder commodity storage is often inadequate (due to lack of appropriate processing and storage facilities), households must resort to other means of coping. Evidence is found to support the hypothesis that agricultural households use enterprise activity as a means of mitigating income shocks. Using panel data from a sample of coffee growers in northwest Tanzania, it is shown that household enterprise ownership goes up by nearly 5 percentage points (or about 12 percent above mean ownership) during coffee price busts. This response is driven by low-investment activities.

Enterprise responses are concentrated among households with lower financial and physical assets, indicating that households with other means of weathering shocks are less likely to choose to use household enterprise. Comparisons of the mean outcomes of enterprises operated by coper households (those operating an enterprise only in periods of low coffee price) and by stayer households (those operating an enterprise throughout the sample period) indicate that the former are less likely to be profitable or hire workers. This relationship, while not a causal one, is not wholly due to the fact that switchers often operate during busts, when local demand for goods and services is weak.

Despite prevailing mitigation mechanisms, the results show that consumption and expenditure are far from fully insured, and so welfare must surely suffer. Secondly, if one takes seriously the idea that there is a distribution of entrepreneurial ability, which governs in part the decision to engage in enterprise along with access to financial resources and production technologies, then the results are consistent with the notion that during price busts, households that otherwise would not have ventured into the enterprise sector are compelled to do so as a means of shock mitigation. For these households, other forms of shock mitigation, such as improved access to savings and credit, may hold more value than enterprise activity.

Achyuta Adhvaryu is an Assistant Professor of Business Economics and Public Policy at the University of Michigan’s Stephen M. Ross School of Business and a Faculty Research Fellow of the National Bureau of Economic Research (NBER); his email address is [email protected]. Namrata Kala is an Assistant Professor of Applied Economics in the MIT Sloan School of Management and a Faculty Research Fellow of the National Bureau of Economic Research (NBER); her email address is [email protected]. Anant Nyshadham is an Assistant Professor of Business Economics and Public Policy at the University of Michigan’s Stephen M. Ross School of Business and a Faculty Research Fellow of the National Bureau of Economic Research (NBER); his email address is [email protected]. The authors thank Prashant Bharadwaj, Eric Edmonds, James Fenske, Dean Karlan, Supreet Kaur, Rocco Macchiavello, Chris Udry, and seminar participants at NBER, Yale, Dartmouth, NEUDC, PACDEV, and RAND for their helpful suggestions. Adhvaryu gratefully acknowledges funding from the National Institutes of Health/Eunice Kennedy Shriver National Institute of Child Health and Human Development (5K01HD071949). A supplementary online appendix is available with this article at The World Bank Economic Review website.

Footnotes

References