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

Transcription

1 Booms, Busts, and Household Enterprise: Evidence from Coffee Farmers in Tanzania Achyuta Adhvaryu Namrata Kala Anant Nyshadham February 2013 Abstract Studies of entrepreneurship in the developing world have struggled to justify the consistent prevalence of low-growth microenterprise and the oft limited adoption and/or utilization of growth opportunities such as microcredit and training in enhanced business practices offered through interventions. Using data from a unique panel of coffee-farming households in Tanzania, we show that intermittent, small-scale enterprise activity plays an important role as a smoothing mechanism in response to shocks to agricultural revenues, the primary source of income for many households in this context. We first verify that negative shocks to the global coffee price significantly decrease farmgate coffee prices, quantities sold, farm revenues and, consequently, household expenditures on food and non-food items. We then show that entrepreneurial entry and time spent in self-enterprise and off-farm employment increase in response to drops in the global coffee price. However, some households persist in enterprise even when coffee prices are high. In fact, these persistent entrepreneurial households increase own labor and employment in enterprise during agricultural windfalls and nearly double performance in terms of survival and profits. Our results suggest that a large subset of households engage in enterprise only intermittently to weather productivity shocks in other sectors and, therefore, stand to gain little from investments in entrepreneurial growth; while other households are more earnestly entrepreneurial, persisting in enterprise, and growing measurably in employment and performance during times of agricultural windfalls. Keywords: household enterprise, coffee production, labor supply, Tanzania Preliminary draft; comments are most welcome. We thank Chris Udry, Dean Karlan, and seminar participants at PACDEV and the Yale Environmental Economics Workshop for their helpful suggestions. Adhvaryu gratefully acknowledges funding from the NIH/NICHD (5K01HD071949) and the Yale MacMillan Center Directors Award. Yale University, achyuta.adhvaryu@yale.edu. web: Yale University, namrata.kala@yale.edu. web: environment.yale.edu/profile/namrata-kala University of Southern California, nyshadha@usc.edu. web: 1

2 1 Introduction A growing literature in economics has sought to understand what barriers prevent microenterprises from expanding and how these barriers may be alleviated. The implied supposition of this literature is that even the smallest entrepreneurs have the desire to expand into largerscale firms, but are constrained by lack of access to financial, human or managerial capital. In this view, interventions that make capital of various types more accessible could facilitate development through small business growth. Recent studies have tested these hypotheses, providing evidence on the effects and sometimes the lack thereof of microcredit, capital infusions, and training in business skills such as management and bookkeeping on the profits and output growth of small and medium enterprises. For instance, in a field experiment in Sri Lanka, de Mel, McKenzie, and Woodruff (2008) find that the return to randomly provided capital amongst microenterprises is higher than the market interest rate and that these returns are even larger for high-ability entrepreneurs. Similarly, Burgess and Pande (2005) find that improved access to banking institutions (which, in theory, provide both credit and savings accounts) in rural India increased output in non-agricultural production. However, Dupas and Robinson (2013) provide experimental evidence from Kenya that access to institutional savings accounts alone can increase productive investments, independent of access to credit. On the other hand, Karlan and Zinman (2010) conduct an experiment in Manila which finds weak average effects of access to microcredit on profits and scale of production of microenterprises. These earlier studies of the relationships between access to financial infrastructure and microenterprise performance and growth is reviewed in-depth in McKenzie (2010). Recent studies have found similarly mixed evidence of the effects of improved access to microcredit on entrepreneurial entry. Crepon et al. (2011) find in Morocco that improved access to credit increased the scale of existing farm and livestock enterprises, but had no effect on the creation of new enterprises. Additionally, amongst non-entrepreneurial households, improved access to credit increased food consumption and durable expenditures. Similarly, Banerjee et al. (2010) find positive effects of access to credit on durable expenditure of entrepreneurial households and non-durable expenditure of non-entrepreneurial households. However, in this context, the results indicate large effects on entrepreneurial entry in the short-run. In a nonexperimental evaluation of one of the largest scale government microfinance initiatives in the world ( Million Baht Village Fund program), Kaboski and Townsend (2012) find no impact of improved access to credit on business starts. Using data from the same context, Nyshadham (2013) finds in a structural approach that dynamic expectations about ability in enterprise are a much stronger determinant of entrepreneurial entry and exit histories of households than are financial constraints. 2

3 Our study is part of a complementary, but relatively nascent literature that has begun to explore ability or skill as another potentially important driver to entrepreneurial entry, performance, and growth. Recent experimental interventions have studied the effects of various training programs on the growth and success of microenterprises. For example, Karlan, Knight, and Udry (2013) develop a model, analogous to the one of entrepreneurial entry and exit in Nyshadham (2013), of decisions along the intensive margin of capital investment with learning about entrepreneurial ability or type through experimentation and employ it to interpret results from a cash grant and managerial capital intervention among tailors in Ghana. The results of training interventions have been arguably equally as confounded as those from the financial infrastructure interventions discussed above. The work to date is well reviewed in McKenzie and Woodruff (2012). Some major takeaways include surprisingly low adoption of training programs, limited implementation of enhanced practices even among adopters, and high attrition among entrepreneurs particularly driven by high frequency switching in and out of the entrepreneurial sector. Overall, the results of previous literature could imply that a large fraction of owners of microenterprises in developing contexts are not, in fact, looking for opportunities to grow their businesses as many policy interventions presuppose. Along these lines, Schoar (2009) posits that entrepreneurs in the developing world can be roughly divided into two categories: transformational and subsistence. The first category is made up of precisely the types of individuals described above: those with the ability and skills necessary to lead large-scale businesses, but who may lack the labor pool or financial resources necessary to expand to their desired capacity. The second category, the subsistence entrepreneurs, is made up of individuals who run tiny operations that do not grow into larger firms but merely provide an alternative employment opportunity to the entrepreneur and potentially their family members (Schoar 2009). Schoar points to evidence that the subsistence category specifically, small, low-growth, familyowned businesses makes up the overwhelming majority of entrepreneurs in most developing country contexts. Moreover, she argues that very few firms transition successfully from subsistence to transformational enterprise. Given the recent, mixed evidence on the effects of capital infusions and management consulting on small-scale enterprise outcomes, it is thus possible that the observed lack of growth beyond subsistence levels in certain types of microenterprises is not the result of market failures necessarily, but rather may be optimal. For example, it is likely that in addition to access to productive capital, entrepreneurs vary substantially in ability, preferences, and outside employment options. This variation may generate heterogeneity in business owners objectives, their economic decisions, such as responses to policy interventions, and ultimately, enterprise outcomes. Given the prevalence and persistence of subsistence entrepreneurship, it is important to understand how, when and why these small-scale household businesses arise. Note many of the 3

4 microfinance interventions discussed above, despite having found weak or confounding effects on business outcomes, have found strong, consistent effects on consumption, suggesting these households are in need of a consumption smoothing mechanism. We propose that the smoothing value of casual enterprises may help to explain their existence, as well as the observed unwillingness of households to grow them into larger-scale businesses. In this study, we provide evidence of this notion by showing that in primarily agricultural households, participation in microenterprise constitutes an important way to weather negative shocks to profitability in the default (farm) sector. This study relates closely to Adhvaryu and Nyshadham (2013) which studies the enterprise activity of households and intra-household labor reallocation in response to acute health shocks to some members of the household. Specifically, in this paper, we find that ownership and time spent in microenterprise in a sample of coffee farmers in Tanzania is strongly inversely related to global coffee prices. Agricultural commodity producers are important to study in this context for several reasons. First, the majority of the global supply of agricultural commodities is from small-holder farming households in low-income countries. Second, these farm households face substantial revenue uncertainty due to fluctuations in global commodity prices, and thus agricultural profitability shocks are salient for them. Third, in developing contexts in which labor markets are often imperfect, offfarm wage employment is not a viable option when faced with low output prices. For all these reasons, microenterprise could play an important smoothing role for these households. We begin our analysis with preliminary descriptions of which types of households own businesses, which types of businesses they own, and which households persist in microenterprise. Specifically, we find that households with heads who are literate and numerate are more likely to own a business in any wave, specifically non-merchant businesses (which tend to require more skill and capital). and are more likely to be entrepreneurial stayers or persistent business owners (own businesses in all 4 waves of the data). On the other hand, households with heads who have some formal education are on average less likely to be entrepreneurial stayers, though not significantly more or less likely to own a business of any type in any particular wave. Access to financial resources is positively related to owning a business in any wave with remittance activity and savings being more associated with non-merchant business and debt and financial stock more associated with merchant businesses. Savings and financial stock are most predictive of persistent entrepreneurship. We then match survey responses of prices, revenues, expenditures, and the labor activities of coffee farming households in a region of northwest Tanzania to the global coffee prices they faced in the last harvest. We first show that the global price and the imputed farmgate price are strongly positively correlated, as are global prices and revenues from the sale of harvested coffee. We show that these revenue shifts generate substantial changes in household food and non-food expenditures. 4

5 Next, we study how global coffee prices affect household entry into the microenterprise sector, measured by ownership of businesses. We find that a one standard deviation rise in the global coffee price significantly decrease the probability of owning a business by roughly 5 percentage points. Specifically, these movements along the extensive margin of enterprise activity pertain entirely to merchant businesses, as opposed to businesses which require more skill and/or capital. We then explore movements along intensive margins using individual-level data on weeks spent in self-employment over the last 6 months. Among households who own businesses, a one standard deviation rise in the global price of coffee drives household members to allocate nearly 3 more weeks to enterprise activity. Again we find that these effects are most significant among households with merchant businesses. These movements along the intensive margin of labor inputs are mirrored in both the number of household members helping with the business and the number of hired workers. However, we do not find significant effects on business capital owned, bought, or sold. Nevertheless, business performance responds significantly with a one standard deviation rise in the global coffee price leading to a 3 month increase in the length of operation of the business, a 23 percentage point increase in the probability of earning positive profits, and a nearly 100% rise in profit level. Finally, we explore heterogeneity in these effects by the demographic characteristics discussed above. We find that households with heads who are both literate and numerate exhibit the strongest effects on the extensive margin (again particularly in merchant businesses), while households with educated heads show weaker effects on business ownership. Effects of interactions with baseline financial variables are weaker, but suggest that positive savings and higher financial stock make households less likely to be entrepreneurial smoothers (that is, less likely to switch into low capital, low skill business in response to coffee price shocks). Taken together, our results are consistent with the notion that household enterprise activity falls into at least two distinct types in this context: smoothing enterprise and persistent enterprise. Smoothing enterprise (entrepreneurship in response to agricultural profitability shocks) is concentrated in the merchant sector and among households that are literate and numerate, but undereducated and active in informal sharing, but have relatively low savings and financial stock. Persistent enterprise, on the other hand, is less likely to be in the low capital, low skill merchant sector and is concentrated among households who are more educated and have higher savings and financial stock. Additionally, entrepreneurial stayer households actually invest more time (both household and hired labor) into their business when agricultural profits are high and their enterprises nearly double in profits and duration of operation. This study contributes to the literature on enterprise in developing countries in at least 3 ways. First, we inform the debate on the existence of heterogeneous types (i.e. preferences, expectations, abilities, access to financial resources) of entrepreneurial households in developing 5

6 contexts and their heterogeneous responses to income shocks. Previous has attempted to define heterogeneous types by observed characteristics in the cross section. This study is able to use optimized behaviors in response to shocks over time to reveal types. Second, we contribute to literature on drivers of entrepreneurial growth. We show that households that persist in enterprise during high agricultural profitability periods allocate more labor hours toward enterprise; hire more labor from outside of the household; operate their businesses for nearly twice as long; and earn nearly double profits. Though agricultural profitability shocks such as the ones studied in this paper will likely have both income and substitution effects, the income effects can to some degree inform expectations about responses to capital infusions and microfinance interventions. Specifically, our results suggest that capital infusions may indeed drive growth, but do so only for a specific households that are best equipped for persistent enterprise. Finally, our results provide preliminary evidence on which types of entrepreneurial households indeed benefit from capital infusions. Our results show that more educated, more financially robust households are more likely to be the persistent entrepreneurs that grow their businesses in times of agricultural windfalls. In some additional checks and results, we emphasize the features of coffee farming that make it appealing in terms of unbiased identification of the effects of agricultural profitability on entrepreneurship. Our main worry in this panel data setting is that households react to coffee prices by changing their area under coffee or, in the extreme, make entry and exit decisions into and out of coffee farming based on the global price. We argue that for coffee farming, entry and exit are not salient issues, since coffee trees take at least three years to produce fruit, and by government regulation, cannot easily be uprooted. Moreover, we then verify in our data that global coffee price fluctuations did not change selection into the coffee grower sample, or affect the area under coffee within the sample. The remainder of the paper is organized as follows. Section 2 describes our data set and construction of important variables. Section 3 presents our empirical strategy and discusses its validity. Section 4 presents results from the empirical tests of the main predictions of the model. Finally, section 5 concludes. 2 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 6

7 of Kagera, interviewed up to four times, from Fall 1991 to January 1994, at 6 to 7 month intervals. The randomized sampling frame was based on the 1988 Tanzanian Census. 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 then were 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 then were randomly sampled. There was moderate attrition from the longitudinal sample. 9.6% of households sampled in wave 1 were lost by wave 4. 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 socio-economic survey following the model of previous World Bank Living Standards Measurement Surveys. The survey covers individual-, household-, and cluster-level data related to the economic livelihoods and health of individuals, and the characteristics of households and communities. Our sample is confined to households who reported harvesting coffee at least once in the survey period ( ), which includes over 80% of the households in the entire sample. We combine the Kagera household survey with data on monthly international coffee prices available with the International Coffee Association. The monthly prices are robusta coffee prices, which is primarily the variety of coffee grown in the Kagera region. Figure 1 shows the graph of monthly prices from , and indicates the prices during the survey period. Prices during the entire survey period ( ) were relatively low compared to the historical average. In the following paragraphs, we outline the variables we use in our analyses. 7

8 Figure I Historical Price Trend ( ) FIGURE 1: ROBUSTA COFFEE PRICE TIME SERIES U.S. Cents/lb jan jan jan jan2010 Date 2.1 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 four waves of the survey however, asked households about their economic and labor activities in the last 6 months. This is because the time lag between waves was about 6-7 months, and the questions were changed to avoid questions about overlapping time periods. In order to estimate the impact of international coffee price fluctuations on the household, we match the outcome variables to the appropriate international price faced by the household at the time when it made decisions regarding labor allocations and microenterprise ownership.since we have information on the month and year in which households were surveyed, we matched the average international price for the time period about which the survey asked. In the first wave, this was the average price for the last 12 months preceding the survey month of the households, and for the subsequent waves, it was the average price for the last 6 months. Thus, 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-August If it was interviewed in any wave other than the first, the price faced by the household is the average international robusta coffee price for the pre- 8

9 ceding 6 months - for instance, if a household was interviewed in September 1993, prices from March 1993-August 1993 would be considered. The average price computed in this manner is about 46 cents/lb, with a standard deviation of about 3.9 cents/lb. Our independent variable of interest is the 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 in the price. 2.2 Household-Level Variables At the household-level, we examine the impact on revenues from coffee, consumption expenditure and microenterprise ownership. Table 1 presents the summary statistics. Area harvested for coffee is on average only about 10% of area harvested by households, but annual revenues from coffee sales comprise about 43% of agricultural revenues for the sample (67% if only households reporting non-zero coffee revenues are included). Thus, it is a significant component of household income. Regarding micro-enterprise ownership, 40% of the households reported owning an enterprise, and about 12% report owning more than one, the maximum being 7. A little over half of these are merchant businesses. We also consider the weeks spent in self-employment in the period of reference, which is conditional on the household owning an enterprise. On average, households spend about 15 weeks in self-employment - merchant business owner households spend the same, and non-merchant business owner households spend 19 weeks. 2.3 Enterprise-level Variables To analyze the operations of the enterprises which continue to operate when the coffee price is high, we use several enterprise-level intensive margin variables. These comprise three categories of enterprise operations - capital assets, labor and performance. The first category includes a binary variable for whether the enterprise owns a capital asset and another binary variable for whether the enterprise bought or sold a capital asset. The majority of enterprises (65%) own a capital asset, although fewer than 15% bought or sold one in the reference period. The labor category comprises the number of household members that helped in the enterprise and the number of hired workers. The number of household members active in the business and the number of hired workers is nearly the same. The performance category consists of three variables - the number of months the enterprise has been operating, the log of business profits, and a binary variable for whether the business had positive profits. The relevant time span for all these variables is that of the survey period mentioned above i.e. the last 12 months for wave 1 and 6 for waves 2,3 and 4. 9

10 3 Empirical Strategy The empirical analysis proceeds in six stages. First, we explore the determinants of business ownership in the pooled sample, by regressing a business ownership dummy on a set of household characteristics meant to capture demographics, financial access, and ability. We estimate the following simple model, in which o is an ownership dummy and h indexes households: o h = α + X h β + ɛ h. (1) Second, we seek to determine the extent to which global coffee prices matter for our sample of coffee farmers. We do this by testing whether global prices are correlated with the farmgate coffee prices that farmers face, and whether the global price affects quantities of coffee harvested, coffee revenues, and household expenditures on food and non-food items. 1 The following model is estimated, for outcome O, price p, and month (θ m ), year (δ y ), and household (µ h ) fixed effects: O hmy = α + βp my + µ h + δ y + θ m + ɛ hmy. (2) As described in the previous section, price p varies at the month x year level. Households surveyed in the same month of a particular wave will thus face the same (retrospective) coffee price; households that happen to have been surveyed in different months of the same wave will face differing prices. Third, we ask how fluctuations in the coffee price affect business ownership. In the coffee grower household sample, we regress an ownership dummy on the coffee price, as well as month, year, and household fixed effects using model specified in equation 5. Fourth, we turn to the individual-level data on business ownership to explore labor supply responses to coffee price movements in the (conditional) sample of individuals who reported owning a business. We regress weeks spent in self-employment in the last 6 months on the coffee price, year and month fixed effects, and individual fixed effects (ζ i ), where individuals are indexed by i: l imy = α + βp my + ζ i + δ y + θ m + ɛ imy. (3) The model specified in equation 4 is estimated on a sample whose unit of observation is thus individual x wave, but is only estimated for individuals who reported owning a business in at least one wave. Fifth, we examine the responses of capital investments, labor, and performance outcomes of 1 Since the farm-gate price that households face is likely endogenously determined (for example, bargaining power of the household or the farming cooperative to which the household belongs could influence farm-gate price), we focus instead on the international price of coffee. Absent stringent price control policies (which were not relevant for our time period in Tanzania), fluctuations over time in the international coffee price should generate exogenous changes in farm-gate prices, and should thus impact agricultural profitability for coffee-growing households. 10

11 enterprises to coffee price movements. For this analysis, we use an unbalanced, enterprise-level panel. The unit of observation is thus enterprise x wave. The specification used is analogous to equation 4, for enteprises indexed by e and enterprise outcome O: O emy = α + βp my + η e + δ y + θ m + ɛ emy. (4) Finally, sixth, we examine heterogeneous business ownership responses to coffee price movements across various household characteristics. The estimate the following interaction specification, for household characteristics X h : o hmy = α + γ (X h x p my ) + βp my + κx h + µ h + δ y + θ m + ɛ hmy. (5) Standard errors in all regressions are clustered at the level of the enumeration cluster. 4 Results 4.1 Which households own a business? We begin by exploring the determinants of household enterprise ownership. We estimate pooled regressions using the whole coffee sample (across all waves). We include the following characteristics of household heads: gender (male dummy); numeracy and literacy dummy (need both for dummy to equal 1); and a dummy for greater than 0 years of grade completion. For the household, we include the following financial characteristics (all dummies): any remittance activity (sent or received); positive savings; positive debt; and above-median financial stock. 2 The coefficient estimates from the regressions are reported in Table 3. In column 1, we see that numeracy/literacy is strongly positively related to business ownership. Households with heads that are both numerate and literate are about 12 percentage points (pp) more likely to engage in entrepreneurial activity, from a mean of about 0.38 (i.e., nearly a 1/3 increase above the mean). Though numeracy/literacy is likely a very noisy predictor of entrepreneurial ability, the result here is consistent with the small but growing literature emphasizing the importance of dimensions of ability in business ownership and success. Household financial variables are highly predictive of ownership, and interestingly, all variables we include positively predict ownership. Remittance activity increases likelihood of business ownership by about 8 pp; positive savings by 11 pp; positive debt by 6 pp; and above median financial stock by more than 8 pp. These results are consistent with the literature on financial deepening and microenterprise growth: financial constraints may play a major role in limiting enterprise development in this context. 2 Please see the data appendix for details on the construction of these and other variables used in analysis. 11

12 Finally, perhaps surprisingly, conditional on numeracy/literacy status, education is not significantly related to ownership: the coefficient is and the standard error band is fairly tight around 0. Households with male heads are slightly more likely to own a business (5 pp), but this relationship is again statistically significant. In columns 2 and 3 of Table 1, we break down enterprises into merchant and non-merchant businesses, and look at correlations separately across the two categories. One should think of the main delineations as size and fixed cost. Merchant businesses are very small, and usually involve very little capital input or fixed costs of operation (e.g., many merchants are vendors that sell fruits, vegetables, and bread on the roadside). Non-merchant businesses are larger and usually involve bigger capital inputs to production (e.g., a construction business or a drug shop with a brick and mortar location). Literacy/numeracy seems to matter for both types of businesses, but in both cases the coefficient estimates are not precise. Education conditional on literacy/numeracy status continues not to play a role in determining ownership. Interestingly, households with male heads are much more likely to own non-merchant businesses, but no such relationship exists for merchant businesses. Remittances and savings matter for non-merchant businesses but not for merchants, whereas for debt and financial stock, the opposite is true. In column 4, we examine the determinants of stayers, that is, those households that operate businesses in all four waves of the panel. We estimate this relationship using household characteristics from wave 1, only using household-level observations from wave 1. The dependent variable is a dummy indicating stayer household. The results here are consistent with those in column 1 with the pooled sample: numeracy/literacy is strongly positively related to stayer status, and financial variables are also all positively related. Education conditional on literacy/numeracy is negatively related to stayer status, suggesting that perhaps households with high innate ability but low grade completion are the most likely to be entrepreneurs in this context. 4.2 Effects of Global Prices on Farming Decisions and Expenditures Next, we test whether global coffee prices actually matter for the coffee farmers in our sample. The general idea is to regress measures of coffee farmgate pricing and production, as well as household expenditures, on the global coffee price. Results are reported in Table 4. We begin by looking at relationship between the global price and the farmgate price, imputed from our transactions data. The results, reported in column 1 of Table 4, demonstrate that the farmgate price is very sensitive to movements in the global price: a one standard deviation (SD) decrease in the global price decreases the farmgate price by 24 percent. 3 3 It bears mentioning that we can only impute the farmgate price for households who had non-zero coffee revenues in the 6-month window prior to survey. 12

13 Next, we examine quantity sold and revenues for coffee. These results, reported in columns 2 and 3, show that both quantity and revenue increase with the global price (by about 30 and 48 percent, respectively, for a one SD change in the global price), consistent with an upwardsloping supply of coffee. Finally, we look at household expenditures, to test the extent to which households are able to smooth consumption in the face of coffee price fluctuations. We find strong evidence against consumption smoothing. Total expenditures decline by nearly 20 percent in the coffee farmer sample for a one SD drop in the coffee price (column 4). We then split expenditures into food and non-food categories, and find substantial changes in expenditures in both corresponding to movements in the coffee price. Food expenditures (column 5 approximately 22 percent) appear to be more elastic than non-food (column 6, approximately 15 percent). 4.3 Household Enterprise Activity and Coffee Price Fluctuations Next, we examine whether global coffee price changes affect the probability of household business ownership. Results are reported in Table 5. Column 1 shows results of a regression of an ownership dummy on the coffee price. We find that a one SD drop in the global price increases the probability of enterprise ownership by about 5 pp, an increase of about 13 percent above the mean. We interpret this finding as strong evidence of countercyclical household enterprise starts in our sample. On average, households are much more likely to engage in enterprise activity during coffee price busts, and shut their businesses during coffee price booms. In columns 2 and 3, we examine ownership of merchant and non-merchant businesses. The results show quite strongly that households are much more likely to smooth using merchant businesses; ownership of non-merchant businesses, on the other hand, does not vary significantly with the coffee price, and the coefficient is 4 times smaller than the impact of coffee price on merchant-type business ownership. 4.4 Labor Supply Effects Conditional on Ownership Next, we examine the impacts of coffee price fluctuations conditional on business ownership, to explore what happens to those households whose businesses stay open during coffee price booms. We begin by looking at time spent in entrepreneurial activity (self-employment) for those individuals who owned a business. Our measure of entrepreneurial activity intensity is the number of weeks the individual spent in self-employment in the past 6 months. The results are reported in Table 6. Column 1 shows self-employment hours conditional on owning at least one business. In this conditional sample, we find a starkly different result on the impact of coffee price movements. During coffee price booms, entrepreneurial activity, conditional on staying open, actually intensifies: weeks in self-employment increase by nearly 13

14 3 for a one SD increase in price. This amounts to an increase of nearly 18 percent over the mean (16 weeks). Columns 2 and 3 report the same effects for the conditional sample of individuals owning merchant and non-merchant businesses, respectively. The coefficient sizes are roughly the same, but the impact is more precisely estimated in the sample of merchant business owners. 4.5 Enterprise Outcomes Conditional on Ownership Next, in the same vein as the labor supply analysis above, we examine households inputs to enterprise and enterprise performance. The results are reported in Table 7. All analysis is done within the unbalanced enterprise-level panel. We first look at capital inputs. Conditional on staying open during a coffee price boom, we find no statistically significant effect of price on the probability of owning or transacting with business assets (columns 1 and 2, respectively). It bears mentioning, however, that these impacts are imprecisely estimated, and thus we cannot reject the null of a fairly large effect on either probability at the conventional levels of confidence. Labor inputs, in contrast, increase significantly with price conditional on enterprise ownership. For a one SD increase in the coffee price, the number of household members helping in the business increases by about 0.2 (column 3) and the number of hired workers increases by more than 0.4 (column 4). Both of these increases are large fractions of their respective means. Performance outcomes also increase during booms conditional on ownership. A one SD increase in the coffee price generates a 3 month increase in the longevity (months of operation out of the past 6 months) of businesses (column 5); a doubling of business profits (column 6); and a 23 pp increase in the probability of positive profits. Overall, the results show resoundingly that for those businesses that remain open during booms, inputs and performance increase substantially. 4.6 Heterogeneous Effects Last, we explore heterogeneity in the enterprise ownership response to coffee price movements across various household characteristics. Table 8 reports results of these interaction specifications. Recall that the main effect from Table 5 showed that households engaged in more entrepreneurial activity during coffee price busts, and shut down enterprise during booms. Interestingly, we find in Table 8 (column 1) that high-ability household heads (those with literacy and numeracy) were less likely to change ownership status with coffee price movements. The impact of coffee price changes does not appear to be differential across financial activity indicators. When we break down ownership into merchant v. non-merchant categories (columns 2 and 3, respectively), we find that ability regulates the relationship between ownership and the coffee 14

15 price only for merchant businesses. Further, higher financial activity and access magnify the impact of coffee price on ownership during price busts, suggesting that financial constraints may hamper the establishment of even these intermittent enterprises. 5 Conclusion Recent literature on drivers of entrepreneurial entry and growth have produced confounded results. The effects of business skills training on growth of existing enterprises appear small and insignificant. Returns to capital among microenterprises appear high in some contexts, but experimental microfinance interventions have produced inconsistent results on entrepreneurial entry and growth of existing enterprises. On the other hand, studies consistently find large effects of improved access to credit on household consumption. We propose that many developing households are subject to income shocks, particularly in agricultural activities, and that they often use microenterprise as a mechanism for weathering these shocks. If, for at least a subset of entrepreneurial households, income-smoothing is the primary motivation of the microenterprise, the aforementioned results of microfinance and business training interventions are less surprising. These households seem to treat microfinance or cash transfers and microenterprise as substitutable mechanisms for smoothing productivity shocks in their ex ante optimal sector of production, largely agriculture. Furthermore, these households are less likely to avail themselves of growth opportunities in the entrepreneurial sector as these businesses do not represent their primary source of income in equilibrium. In particular, the subsistence entrepreneur might be constrained in terms of growth by ability more than credit. That is, the households who switch frequently in and out of the entrepreneurial sector or allocate on average a small fraction of the time and resources towards their microenterprises do not likely represent the high ability entrepreneurs. In an effort to differentiate intermittent entrepreneurial households from persistent entrepreneurial households, this paper explores how household enterprise activity responds to shocks to agricultural profitability. Specifically, we study how global coffee prices affect household entry into the microenterprise sector, measured by ownership of businesses. We find that a rise in the global coffee price significantly decreases the probability of owning a business. We also find that among households who own businesses, a rise in the global price of coffee drives household members to allocate more productive time to enterprise activity. We find that these effects along the extensive and intensive margins are most significant with respect to low skill, low capital merchant businesses. These movements along the intensive margin of labor inputs are mirrored in both the number of household members helping with the business and the number of hired workers. Though we do not find significant effects on business capital owned, bought, or sold, business performance responds significantly in terms of length of operation and profits. 15

16 Finally, we explore heterogeneity in these effects by the demographic characteristics such as literacy, numeracy and education of household head and the household s access to both formal and informal financial resources. We find that households with heads who are both literate and numerate exhibit the strongest effects on the extensive margin (again particularly in merchant businesses), while households with educated heads show weaker effects on business ownership. Effects of interactions with baseline financial variables are weaker, but suggest that positive savings and higher financial stock make households less likely to be entrepreneurial smoothers (that is, less likely to switch into low capital, low skill business in response to coffee price shocks). Taken together, our results are consistent with the notion that household enterprise activity falls into at least two distinct types in this context: intermittent, smoothing enterprise and persistent enterprise. Smoothing enterprise (entrepreneurship in response to agricultural profitability shocks) is concentrated in the merchant sector and among households that are literate and numerate, but undereducated and active in informal sharing, but have relatively low savings and financial stock. Persistent enterprise, on the other hand, is less likely to be in the low capital, low skill merchant sector and is concentrated among households who are more educated and have higher savings and financial stock. Additionally, entrepreneurial stayer households actually invest more time (both household and hired labor) into their business when agricultural profits are high and their enterprises nearly double in profits and duration of operation. These results provide suggestive evidence that, Indeed, allocating public resources toward improving access to credit for high ability, growth-oriented entrepreneurs might be a high return endeavor, both in terms of growth of these enterprises and from a general welfare-enhancing perspective. However, identifying this high return subset of entrepreneurs is, then, of the utmost importance. To the best of our knowledge, very few studies have attempted to describe microenterprise owners in terms of their ability, risk preferences, motivations, expectations, etc. de Mel, Mckenzie, and Woodruff (2008) describe microenterprise owners in from a study sample in Sri Lanka with respect to some of these dimensions, but are only able to associate these characteristics with enterprise performance and scale in the cross-section. Our study is able to categorize entrepreneurial households by their heterogeneous optimized enterprise responses to shocks over time, providing some preliminary contributions to the literature s understanding of which types of households represent persistent, high growth potential entrepreneurs and which types of households are more likely to be intermittent, smoothing entrepreneurs. Finally, though recent business training interventions have produced little in the way of measurable success, they could be quite valuable in converting the marginal subsistence entrepreneur into a transformational entrepreneur with growth potential. Both describing this subset of the population in greater detail and testing this hypothesis are important endeavors for future studies. 16

17 References [1] Adhvaryu, Achyuta and Anant Nyshadham Health, Enterprise, and Labor Complementarity. Working Paper. [2] Banerjee, Abhijit, Esther Duflo, Rachel Glennerster, and Cynthia Kinnan The miracle of microfinance? Evidence from a randomized evaluation. Working Paper. [3] Burgess, Robin and Rohini Pande Do rural banks matter? Evidence from the Indian Social Banking Experiment. American Economic Review, vol. 95, no. 3: [4] Crepon, Bruno, Florencia Devoto, Esther Duflo, and William Pariente Impact of microcredit in rural areas of Morocco: Evidence from a Randomized Evaluation. Working Paper. [5] de Mel, Suresh, David McKenzie, and Christopher Woodruff Returns to Capital in Microenterprises: Evidence from a Field Experiment. Quarterly Journal of Economics, vol. 123, no. 4: [6] de Mel, Suresh, David McKenzie, and Christopher Woodruff Who are the microenterprise owners? Evidence from Sri Lanka on Tokman v. de Soto. International Differences in Entrepreneurship. Joshua Lerner and Antoinette Schoar (Eds.): pp [7] Dupas, Pascaline and Jonathan Robinson Savings Constraints and Microenterprise Development: Evidence from a Field Experiment in Kenya. American Economc Journal: Applied Economics. Vol. 5, no. 1: [8] Kaboski, Joseph and Robert Townsend The Impact of Credit on Village Economies. American Economic Journal: Applied Economics. forthcoming [9] Karlan, Dean and Martin Valdivia Teaching Entrepreneurship: Impact of Business Training on Microfinance Clients and Institutions. Review of Economics and Statistics, vol. 93, no. 2: [10] Karlan, Dean and Jonathan Zinman Expanding Credit Access: Using Randomized Supply Decisions to Estimate the Impacts. Review of Financial Studies, vol. 23, no. 1: [11] Karlan, Dean, Ryan Knight, and Christopher Udry Hoping to Win, Expected to Lose: Theory and Lessons on Micro Enterprise Development. Working Paper. [12] McKenzie, David Impact Assessments in Finance and Private Sector Development: What have we learned and what should we learn? World Bank Research Observer Vol. 25, no. 2: [13] McKenzie, David and Christopher Woodruff What are we learning from business training and entrepreneurship evaluations around the developing world? working paper. [14] Nyshadham, Anant Learning about Comparative Advantage in Entrepreneurhsip: Evidence from Thailand. working paper. 17

18 [15] Schoar, Antoinette The Divide between Subsistence and Transformational Entrepreneurship. Innovation Policy and the Economy, Vol. 10, No. 1: pp

19 A Construction of Variables The following list describes the construction of variables used in analysis. Please note that since the survey interviewed households about their decisions and outcomes in the 12 months preceding the interview in the first wave, and six months preceding the interview in the second, third and fourth wave, unless otherwise mentioned, all household-level variables are defined for these respective time-periods. Furthermore, with the exception of the regression containing the variable Coffee Grower as the dependent variable, all other regressions are conditional on the household having reported harvesting coffee in at least one of the four waves. A.1 Coffee Price Variabes Robustalag: a household-level lagged international robusta coffee price. For the first wave, it is the mean price of the twelve months preceding the interview of the household, and for the remaining waves, it is the mean price of the six months preceding the household interview. Since different households were interview at different times, it varies across households, though households interviewed in the same month and year will have identical values of robustalag. This is done to align the outcome variables, which were asked for the previous twelve months in wave 1 and previous six months in the other waves, with the mean coffee price prevailing in the time-interval when the outcomes were determined. The monthly-level international robusta prices were obtained from the International Coffee Organization(ICO). Price/SD(Price): a primary independent variable. It equals robustalag divided by the standard deviation of robustalag. Thus, the coefficients on this variable in the regressions indicate the marginal impact of a one standard deviation increase in robustalag. A.2 Enterprise Ownership 1(Household Owns a Business): equals 1 if the the household reported to owning at least one enterprise, 0 otherwise. 1(Household Owns a Merchant Business): equals 1 if the household reported owning at least one enterprise that undertook trading or other informal non-farm self-employment. 1(Household Owns a Non-Merchant Business): equals 1 if the household reported owning an at least enterprise of the categories that require semi-skilled and skilled work. These comprise stall keeping, shopkeeping, restaurant owner, garage owner, bus driver, blacksmith, plumber, carpenter, tailor, repair work, mechanic, mason, painter, hair dresser, shoemaker, butcher, handicrafts, photographer, and doctor. 19