Booms, Busts, and Household Enterprise: Evidence from

Coffee Farmers in Tanzania∗

Achyuta Adhvaryu†

Namrata Kala‡

Anant Nyshadham§

March 2013

Abstract

Studies of entrepreneurship in the developing world have struggled to explain the highprevalence of small, low-growth enterprises, and the limited impacts of financial and train-ing interventions on business expansion. Using data from a unique panel of coffee farmerhouseholds in Tanzania, we demonstrate an important role for microenterprise activity asa means of mitigating the effects of income shocks. We first verify that global coffee pricesmatter for these households, through their effects on farmgate prices, quantity of coffee sold,farm revenues and, consequently, household expenditures. We then show that householdsare more likely to engage in enterprise activity during coffee price busts. Those householdswhose businesses stay open during booms, however, actually increase input intensity andreap large rewards in terms of business survival and profits. Our results suggest that manyagricultural households only intermittently engage in enterprise activity, as a way to weatherproductivity or revenue shocks in their principal sector. While some of these household busi-nesses may grow when financial or managerial constraints are relaxed, many find little valuein entrepreneurship beyond as a means of smoothing.

Keywords: microenterprise, coffee, agricultural shocks, Tanzania

∗Preliminary draft; comments are most welcome. We thank Chris Udry, Dean Karlan, and seminar participantsat PACDEV and the Yale Environmental Economics Workshop for their helpful suggestions. Adhvaryu gratefullyacknowledges funding from the NIH/NICHD (5K01HD071949) and the Yale MacMillan Center Directors Award.†Yale University, e-mail: achyuta.adhvaryu@yale.edu. web: www.yale.edu/adhvaryu‡Yale University, e-mail: namrata.kala@yale.edu. web: environment.yale.edu/profile/namrata-kala§University of Southern California, e-mail: nyshadha@usc.edu. web: www.anantnyshadham.com

1

1 Introduction

A growing body of work in economics seeks to understand what barriers prevent small andmedium-scale enterprises from expanding, and how these barriers may be alleviated. The im-plied supposition of many studies in this literature is that even the smallest entrepreneurs havethe desire to expand into larger-scale firms, but are constrained by lack of access to financial,human, or managerial capital. In this view, interventions that make various types of capitalmore accessible could facilitate economic development through small business growth.

Credible evidence from randomized controlled trials has demonstrated the effects–or oftenthe lack thereof–of microcredit, capital provision, input subsidies, and the like. Impacts on laborand capital inputs, revenues, and profits of small and medium enterprises have been evaluated.Experimental studies of increased access to credit or financial capital have found heterogeneouseffects on consumption and expenditure on durables by enterprise activity of the household,but mixed evidence of effects on entrepreneurial entry and enterprise growth (e.g. de Mel,McKenzie, and Woodruff (2008); Crepon et al. (2012); Banerjee et al. (2012)).1

Some very recent work has begun to explore ability or skill (human and managerial capi-tal) as another potentially important driver for entry into entrepreneurship, performance, andgrowth. The typical study in this literature evaluates the impacts of training in business skills,such as management or bookkeeping. The results of these studies are mixed: some find large,positive effects on existing enterprise performance (e.g. Karlan and Valdivia (2011); Bruhn etal. (2012)), while others find insignificant or even negative effects (e.g. de Mel, McKenzie,and Woodruff (2012); Karlan, Knight, and Udry (2013)). Completed studies in this literatureto date are reviewed in McKenzie and Woodruff (2012). As mentioned in this review, somecommon issues with these studies include surprisingly low adoption of training programs; lim-ited implementation of recommended practices, even among adopters; and high attrition rates,particularly driven by high frequency switching in and out of the entrepreneurial sector.

Most of these studies work with a sample of existing enterprises, so they cannot explainthe entrepreneurial entry decision, and are limited in their ability to deal with the frequentswitching of household participation in enterprise.2 In this study, we show that agriculturalhouseholds frequently switch in and out of enterprise, and that part of this switching is drivenby shocks to the farm sector. Specifically, we show, in a sample of Tanzanian coffee farmers,that when the coffee price is low, households are more likely to operate small-scale businesses;during coffee price “booms,” households shut these businesses and devote their time to tendingtheir farms.

Agricultural commodity producers are important to study in this context for several reasons.

1See Kaboski and Townsend (2012) for a non-randomized study of improved access to credit. For a review ofearlier studies of improved access to financial resources on enterprise outcomes see McKenzie (2010).

2Some important recent exceptions are Crepon et al. (2012), Banerjee et al. (2012), and de Mel et al. (2012).

2

First, the majority of the global supply of agricultural commodities is from small-holder farminghouseholds in low-income countries. Second, these farm households face substantial revenueuncertainty due to fluctuations in global commodity prices, and thus agricultural profitabilityshocks are salient for them. Third, in developing country contexts in which labor markets areoften imperfect, off-farm wage employment is not a viable option when faced with low outputprices. For all these reasons, microenterprise could play an important smoothing role for thesehouseholds.

We begin our analysis with preliminary descriptions of the types of households who ownbusinesses, which types of businesses they own, and the types of households who persist in mi-croenterprise. Households headed by literate and numerate individuals are more likely to owna business (specifically non-merchant businesses, which tend to require more skill and capital).These households are also more likely to be persistent business owners, or “entrepreneurialstayers,” which we define as households who own businesses in all 4 rounds of the panel sur-vey. Access to financial resources is positively related to owning a business in any wave, withremittance activity and savings being more associated with non-merchant business, and debtand financial stock more associated with merchant businesses. Savings and financial stock aremost predictive of persistent entrepreneurship.

We match survey responses of prices, revenues, expenditures, and the labor activities ofcoffee farming households in a region of northwest Tanzania to the global coffee prices theyfaced in the last harvest. We first show that the global price and the imputed farm-gate priceare strongly positively correlated, as are global prices and household revenues from the sale ofharvested coffee. We show that these revenue shifts generate substantial changes in householdfood and non-food expenditures.

Next, we study how global coffee prices affect household entry into the enterprise sector,measured by ownership of businesses. We find that a one standard deviation rise in the globalcoffee price significantly decrease the probability of owning a business, by roughly 5 percentagepoints. Specifically, these movements along the extensive margin of enterprise activity pertainentirely to merchant businesses, as opposed to businesses which require more skill and/or cap-ital.

We then explore movements along the intensive margin, in the allocation of labor, capital,and enterprise performance. Among households who own businesses, a one standard deviationrise in the global price of coffee drives household members to allocate nearly 3 more weeks toenterprise activity in a given 6 or 12-month period. Once again, we find that these effects aremost significant among households with merchant businesses.

This response of labor inputs is mirrored in both the number of household members helpingwith the business and the number of hired workers. However, we do not find significant effectson business capital owned, bought, or sold. Business performance responds significantly with a

3

one standard deviation rise in the global coffee price leading to a 3 month increase in the lengthof operation of the business, a 23 percentage point increase in the probability of earning positiveprofits, and a nearly 100% rise in profit level.

Finally, we explore heterogeneity in these effects by the demographic characteristics dis-cussed above. We find that households where the head of the household is both literate andnumerate exhibit the strongest effects on the extensive margin (again particularly in merchantbusinesses), while households with educated heads show weaker effects on business owner-ship. Effects of the interactions with baseline financial variables are weaker, but suggest thatpositive savings and higher financial stock make households less likely to be entrepreneurialsmoothers (that is, less likely to switch into low capital, low skill business in response to coffeeprice shocks).

Coffee farming has features that enable us to identify unbiased estimates of the effects ofagricultural profitability on entrepreneurship. A main concern for identification is that house-holds react to coffee prices by changing the area farmed under coffee or in the extreme, makeentry and exit decisions into and out of coffee farming based on the global price. We argue thatfor coffee farming, entry and exit are not salient issues, since coffee trees take at least three yearsto produce fruit, and consequently households cannot scale up or down the area under coffee ina short period of time in response to global price fluctuations. Moreover, we verify in our datathat global coffee price fluctuations did not change selection into the coffee grower sample, oraffect the area under coffee within the sample.

Taken together, our results are consistent with the notion that household enterprise activ-ity falls into at least two distinct types in this context: intermittent, smoothing enterprise andpersistent enterprise. Smoothing enterprise (entrepreneurship in response to agricultural prof-itability shocks) is concentrated in the merchant sector and among households that are literateand numerate, but undereducated and active in informal risk sharing, and have relatively lowsavings and financial stocks. Persistent enterprise, on the other hand, is less likely to be in thelow capital, low skill merchant sector and is concentrated among households who are moreeducated and have higher savings and financial stocks. Additionally, entrepreneurial stayerhouseholds (those who own at least one enterprise in all 4 waves of the data) actually investmore time (both household and hired labor) into their business when agricultural profits arehigh, and their enterprises nearly double in profits and duration of operation in a given year.

Overall, the results of previous literature on access to both financial and managerial capitalsuggest, perhaps, that a sizeable fraction of owners of microenterprises in developing contextsare not, in fact, looking for opportunities to grow their businesses, as many policy interventionspresuppose. It is notable that many of the microfinance interventions discussed above, despitehaving found weak or confounding effects on business outcomes, have found strong effects onconsumption and expenditure, suggesting that these households are in need of a smoothing

4

mechanism. We propose in this paper that the smoothing value of casual enterprises may helpexplain their existence, as well as the observed unwillingness of households to grow them intolarger-scale businesses.

We provide evidence of this notion by showing that in primarily agricultural households,participation in microenterprise constitutes an important way to weather negative shocks toprofitability in the agricultural sector, which in this context is the default production sector.This study relates closely to Adhvaryu and Nyshadham (2013), which explores the enterpriseactivity of households and intra-household labor reallocation in response to acute health shocksto some members of the household.

This paper contributes to the literature on entrepreneurship in developing countries in threeways. First, we inform the debate on the existence of heterogeneous types of entrepreneurialhouseholds (in terms of preferences, expectations, abilities, and access to financial resources),and their heterogeneous responses to income shocks. Previous studies (e.g. de Mel, McKenzie,and Woodruff (2008)) have attempted to define heterogeneous types by observed characteristicsin the cross section. This study is able to use optimized behaviors in response to shocks overtime to reveal the types of entrepreneurial households. 3

Second, we contribute to the literature on the drivers of entrepreneurial growth. We showthat in a given year, households that persist in enterprise during periods of high global coffeeprices and therefore high agricultural profitability allocate more labor hours toward enterprise,hire more labor from outside of the household, operate their businesses for nearly twice aslong, and earn nearly double profits. Though agricultural profitability shocks such as the onesstudied in this paper will likely have both income and substitution effects, the income effectscan to some degree inform expectations about responses to capital provision and microfinanceinterventions. Our results suggest that capital infusions may indeed drive growth, but will doso only for the types of households who are best equipped for persistent enterprise. Finally, weprovide preliminary evidence on which types of entrepreneurial households benefit from capitalinfusions. Our results show that “higher ability” (literate and numerate) and more financiallyrobust households are more likely to be the persistent entrepreneurs that grow their businessesin times of agricultural windfalls.

The remainder of the paper is organized as follows. Section ?? reviews the previous litera-ture. Section 2 develops a simple model to formalize our intuition about household enterpriseownership and operation during booms and busts. Section 3 describes our data set and con-struction of important variables. Section 4 presents our empirical strategy and discusses itsvalidity. Section 5 presents results from the empirical tests of the main predictions of the model.

3This approach is similar in spirit to Nyshadham (2013) which uses optimized histories of household enterpriseresponses to productivity shocks in Thailand to test the roles of ability and financial constraints in determiningentrepreneurial entry.

5

Finally, section 6 concludes.

2 Model

We develop a simple model meant to capture some salient features of households’ decisionsto engage in entrepreneurship. Households in our model are “primarily” farmers–a fact thatis modeled through a land endowment and land-labor complementarity–but can also chooseto operate a business. Households differ in entrepreneurial ability, which drives differences inenterprise sector participation. We show that when farm prices are low, households are morelikely to participate in the business sector; when prices are high, they shut their businesses tospend more time on the farm. High-ability tend to stay in business even when the farm price ishigh–we show that this ability-based selection implies that for those “stayers,” business inputs,revenues, and profits might actually go up on average when the farm price is high.

2.1 Setup

Consider a household whose preferences for consumption are represented by the utility func-tion u(c). Suppose u is differentiable and increasing in consumption. The household generatesincome by participating in up to two sectors: farm production and (non-farm) enterprise. Thefarm production function, f , takes labor (lf ) and land (or in our context, trees, T ) as inputs. Weassume that T is fixed in the short run.4 The enterprise production function, e, takes as inputslabor (le) and capital (k); the latter can be rented at rate r. We assume enterprise production isscaled by an ability parameter α ∈ (0,∞), which is meant to capture entrepreneurial or manage-rial skill. For parsimony’s sake, we assume leisure is not valued; the household’s time constraintis thus lf + le = Ω, where Ω is the endowment of time. The price of farm output is pf and theprice of business output is pe. Given these parameters, the household’s utility maximizationproblem is:

maxc,lf ,le,k u(c) s.t. (1)

c ≤ pff(lf , T ) + pee(le, k, α)− rk, (2)

lf + le = Ω. (3)

We render this general problem slightly more specific to be able to formalize the impactsof farm (coffee) price booms and busts on entrepreneurship. Suppose that both production

4This is in keeping with much of the sub-Saharan African context: land markets are thin, if at all existent. Inour context, households during the early 1990s in Tanzania were forbidden by law from uprooting coffee trees, andnewly planted trees take 3-5 years to grow to maturity before producing coffee cherries. We verify in our results thatacreage under coffee does not change with the coffee price.

6

functions are Cobb-Douglas, with weights on farm and enterprise labor equal to γ and δ, re-spectively:

f(lf , T ) := lγfT1−γ (4)

e(le, k, α) := αlδek1−δ. (5)

2.2 Optimal Allocation of Labor and Capital

2.2.1 Interior Solution

At an interior solution to the above maximization problem, the necessary first order conditionsfor lf and k must be satisfied (we denote solutions to the maximization problem by ∗):5

lf : pfγ

(T

l∗f

)1−γ

= peαδ

(k∗

Ω− l∗f

)1−δ

(6)

k : peα (1− δ)(

Ω− l∗fk∗

)δ= r (7)

Combining equations 6 and 7, with some tedious algebra we can show that at an interiorsolution,

l∗f = T

(pfγ(1− δ)

δr

(r

peα(1− δ)

) 1δ

) 11−γ

(8)

l∗e = Ω− T

(pfγ(1− δ)

δr

(r

peα(1− δ)

) 1δ

) 11−γ

(9)

k∗ =

(peα(1− δ)

r

) 1δ

Ω− T

(pfγ(1− δ)

δr

(r

peα(1− δ)

) 1δ

) 11−γ . (10)

The above expressions for optimal allocations of capital and labor across sectors have someintuitive implications. For example, as T , the endowment of trees, increases, farm labor shouldincrease (because of land-labor complementarity in the farm production function), and laborand capital investment in enterprise should decrease. As entrepreneurial ability α increases,time spent in the entrepreneurial sector and capital invested in the household business increase,while time spent on the farm decreases.

5Note that only two first order conditions are necessary once we substitute for consumption using the budget andtime constraints.

7

2.2.2 Corner Solution

Of course, the endowment of land (trees) T , combined with the land-labor complementarityinherent in the Cobb-Douglas form of the farm production function, makes salient the possibilityof a corner solution, in which the household works only in the farm sector, and does not operatea business (despite free entry into the enterprise sector). In this corner solution, the householddevotes all its time to farm labor (lf = Ω), and no time to business activity (le = 0). Thus, utilityat the corner is u(pff(Ω, T )).

2.2.3 The Entrepreneurial Decision

The household, therefore, engages in entrepreneurship if and only if

u (pff(Ω, T )) < u(pff(l∗f , T ) + peαe(l

∗e , k∗)− rk∗

). (11)

Equivalently, since u is monotonically increasing in c, the household engages in enterpriseactivity if and only if

pff(Ω, T ) < pff(l∗f , T ) + peαe(l∗e , k∗)− rk∗. (12)

2.3 Comparative Statics

Our main purpose is to investigate how the entrepreneurial decision and the optimal allocationsof capital and labor conditional business ownership change during coffee price booms (in termsof model parameters, increases in pf ). We will also use the model to study how coffee pricebooms and busts affect entrepreneurial choices along the ability distribution.

We first compute the comparative statics for the optimized quantities l∗f , l∗e , and k∗. Fromequations 8, 9, and 10, respectively, it is evident that for each household, an increase in pf wouldgenerate an increase in l∗f and decreases in l∗e and k∗.6

2.3.1 Who chooses to operate a business?

Suppose that each household i receives a draw αi from an ability distribution A on the interval(0, 1), with associated probability density function a. Household i decides to engage in theenterprise sector if and only if

∆Γi := pff(Ω, T )− pff(l∗f , T )− peαie(l∗e , k∗) + rk∗ < 0. (13)

6The proof is quite simple and so is omitted.

8

The following lemma states that there exists an ability type cutoff such that all householdswith ability above the cutoff should participate in the enterprise sector, and all below shouldnot.

Lemma 1 There exists an ability cutoff α such that αi > α⇔ ∆Γi < 0.

Proof. Showing the existence of a cutoff value α as defined above is equivalent to demonstratingthat ∆Γ is a monotonic function of α. Differentiating the left-hand side of the inequality inequation 13 with respect to α and rearranging terms, we obtain:

−pee︸ ︷︷ ︸X4

+∂l∗e∂α

(pff1 − peαe1)︸ ︷︷ ︸X5

−∂k∗

∂α(r − peαe2)︸ ︷︷ ︸

X6

. (14)

X4 is negative, and, as above, X5 and X6 are equivalent to the necessary first order condi-tions, which must equal 0 at an optimum. Thus, the derivative above is negative, implying that∆Γ is strictly decreasing in α. Therefore, there exists some α such that αi > α⇔ ∆Γi < 0.

Thus, holding other parameter values fixed, the model makes the intuitive prediction thathigher ability types will choose into entrepreneurship, while lower ability types will prefer thecorner solution, devoting all their time to farm production.

2.3.2 Who “drops out” and who “stays in” during coffee price booms?

To understand what happens to entrepreneurial choice across the distribution of types duringcoffee price booms, we must determine how α changes in response to an increase in pf .

We begin with the following lemma, which states that for each αi, ∆Γi–the value of choosingfarming alone over farming and enterprise participation–is increasing in the coffee price.

Lemma 2 ∆Γi is increasing in pf .

Proof. Differentiating ∆Γi with respect to pf and rearranging terms, we obtain:7

f(Ω, T )− f(l∗f , T )︸ ︷︷ ︸X1

+∂l∗f∂pf

(peαie1 − pff1)︸ ︷︷ ︸X2

−∂k∗

∂pf(peαie2 − r)︸ ︷︷ ︸

X3

. (15)

The term X1 above is positive, since f is increasing in labor and Ω > l∗f . The terms X2 andX3 must equal 0 at an optimum, since l∗f , k∗, and l∗e are defined by exactly these necessary first

7f1, e1, and e2 denote, respectively, the partial derivatives of the farm production function with respect to labor,the enterprise production function with respect to labor, and the enterprise production function with respect tocapital.

9

order conditions (see equations 6 7). The derivative of ∆Γi is thus positive.

The following proposition states that the ability cutoff rises during coffee price booms.

Proposition 3 α is increasing in pf .

Proof. Take p1f < p2

f , and define α1 and α2 as the solutions to the equations ∆Γ|p1f (α1) = 0

and ∆Γ|p2f (α2) = 0, respectively. By lemma 2, for each αi, ∆Γi|p2f > ∆Γi|p1f . In particular,∆Γ|p2f (α1) > ∆Γ|p2f (α1) = 0. Now, since ∆Γ|p2f (α2) = 0 by construction, and ∆Γ is decreasing inα by lemma 1, it must be that α1 < α2.

Since the ability cutoff α goes up during price booms, the mass of households participatingin entrepreneurship declines. So, we learn from the above proposition that during price booms,there is, on average, exit from the enterprise sector, but the entrepreneurs who keep their busi-nesses open are the higher-ability ones.

2.3.3 Labor, Capital, Revenues, and Profits for “Stayers”

The final question we study is what happens in this model to inputs, revenues, and profits forthose (high-ability type) households who continue to own businesses during price booms. Webegin with inputs (labor and capital), and then study revenues and profits.

Enterprise labor: We average the expression for labor inputs in 9 over the part of the abilitydistribution that participates in enterprise (αi > α). The average l∗e |α is equal to

l∗e |α =1

1−A(α)

∫ ∞α

l∗ea(α)dα. (16)

Differentiating equation 16 with respect to pf , we obtain:

∂l∗e |α∂pf

=a(α)

(1−A(α))2

∂α

∂pf

∫ ∞α

l∗ea(α)dα+1

1−A(α)

∂

∂pf

∫ ∞α

l∗ea(α)dα. (17)

The first term in the equation above equals a(α)1−A(α)

∂α∂pf

l∗e |α. The second term above, by Leib-

niz’s rule, equals 11−A(α)

(∫∞α

∂l∗e∂pf

a(α)dα− l∗e |αa(α) ∂α∂pf

). Plugging these expressions in above

and rearranging terms, ∂l∗e |α∂pf

can be expressed as:

∂l∗e |α∂pf

=1

1−A(α)

(∫ ∞α

∂l∗e∂pf

a(α)dα+∂α

∂pfa(α)

(l∗e |α − l∗e |α

)). (18)

10

The first term within parentheses in the equation above is negative, since for every α, ∂l∗e∂pf

<

0. The second term is positive, since ∂α∂pf

> 0 by proposition 3; a(α) > 0; and l∗e |α − l∗e |α > 0

because l∗e is strictly increasing in α.Equation 18 thus reveals two competing forces pushing average labor inputs for “stayers”

in opposite directions. First, when pf increases, for given αi > α, l∗e decreases, because thereis a pressure, arising from the additional profitability of farm production, to put more laborinto the farm sector (first term within parentheses). Second, when pf increases, higher abil-ity entrepreneurs are the ones who stay in, and those households, given their higher returnsto entrepreneurship, spend more time in enterprise than lower ability households, causing anupward pressure on average labor inputs into enterprise (second term).

The overall sign of ∂l∗e |α∂pf

is ambiguous, and depends on the size of relative size of the equa-tion’s various components. For example, if enterprise labor supply is highly elastic with respectto the coffee price, average labor for “stayers” may decrease. On the other hand, if enterpriselabor is steeply increasing in ability, or if the cutoff α is highly responsive to the coffee price, theopposite – an increase in average labor for “stayers” – is possible.

Capital inputs, revenues, and profits: Following an analogous derivation, we can express thechange in average capital inputs, revenues, and profits (Π := pee(l, k, α) − rk) for “stayers”during price booms as follows:

∂k∗|α∂pf

=1

1−A(α)

(∫ ∞α

∂k∗

∂pfa(α)dα+

∂α

∂pfa(α)

(k∗|α − k∗|α

))(19)

pe∂e∗|α∂pf

=pe

1−A(α)

(∫ ∞α

(e1∂l∗e∂pf

+ e2∂k∗

∂pf

)a(α)dα+

∂α

∂pfa(α) (e∗|α − e∗|α)

)(20)

∂Π∗|α∂pf

= pe∂e∗|α∂pf

− r∂k∗|α

∂pf. (21)

The same tradeoff discussed earlier for average labor inputs applies to these equations. Notethat if average labor and capital inputs both increase during price booms, it must be that averagerevenues also goes up, given that both enterprise labor and capital are monotonically increasingin ability. However, average profits need not go up, since the sign of the change in profitsdepends on both the change in average revenues and the change in average capital.

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 engagedin producing bananas and coffee in the north, and rain-fed annual crops (maize, sorghum, and

11

cotton) in the south. The Kagera Health and Development Survey (KHDS) was conducted bythe World Bank and Muhimbili University College of Health Sciences. The sample consists of816 households from 51 clusters (or communities) located in 49 villages covering all five districtsof Kagera, interviewed up to four times, from Fall 1991 to January 1994, at 6 to 7 month inter-vals. 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 (orcommunities) were stratified based on agro-climactic zone and mortality rates and then wererandomly 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 12months before enumeration, and then were randomly sampled. There was moderate attritionfrom the longitudinal sample. 9.6% of households sampled in wave 1 were lost by wave 4. How-ever, to preserve balancing across health profiles in the sample, lost households were replacedwith randomly selected households from a sample of predetermined replacement householdsstratified by sickness. KHDS is a socio-economic survey following the model of previous WorldBank Living Standards Measurement Surveys. The survey covers individual-, household-, andcluster-level data related to the economic livelihoods and health of individuals, and the charac-teristics of households and communities. Our sample is confined to households who reportedharvesting coffee at least once in the survey period (1991-1994), which includes over 80% of thehouseholds in the entire sample.

We combine the Kagera household survey with data on monthly international coffee pricesavailable with the International Coffee Association. The monthly prices are robusta coffeeprices, which is primarily the variety of coffee grown in the Kagera region. Figure 1 showsthe graph of monthly prices from 1980-2010, and indicates the prices during the survey period.Prices during the entire survey period (1991-1994) were relatively low compared to the historicalaverage. In the following paragraphs, we outline the variables we use in our analyses.

3.1 Price Lag Variable

The first wave of the survey asked households about their economic and labor activities in the12 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 becausethe time lag between waves was about 6-7 months, and the questions were changed to avoidquestions about overlapping time periods. In order to estimate the impact of international cof-fee price fluctuations on the household, we match the outcome variables to the appropriateinternational price faced by the household at the time when it made decisions regarding laborallocations and microenterprise ownership.Since we have information on the month and yearin which households were surveyed, we matched the average international price for the timeperiod about which the survey asked. In the first wave, this was the average price for the last

12

FIGURE 1: ROBUSTA COFFEE PRICE TIME SERIES

Figure I

Historical Price Trend (1980 – 2010)

050

100

150

200

U.S

. Cen

ts/lb

.

01jan1980 01jan1990 01jan2000 01jan2010Date

13

12 months preceding the survey month of the households, and for the subsequent waves, itwas the average price for the last 6 months. Thus, if a household was interviewed in wave 1in September 1991, the price faced by the household is the average international robusta coffeeprice from September 1990-August 1991. 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-ceding 6 months - for instance, if a household was interviewed in September 1993, prices fromMarch 1993-August 1993 would be considered. The average price computed in this manner isabout 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 deviationover the survey period. The coefficient on this variable is the marginal effect of a one standarddeviation in the price.

3.2 Household-Level Variables

At the household-level, we examine the impact on revenues from coffee, consumption expen-diture and microenterprise ownership. Table 1 presents the summary statistics. Area harvestedfor coffee is on average only about 10% of area harvested by households, but annual revenuesfrom coffee sales comprise about 43% of agricultural revenues for the sample (67% if only house-holds’ reporting non-zero coffee revenues are included). Thus, it is a significant component ofhousehold income.

Regarding micro-enterprise ownership, 40% of the households reported owning an enter-prise, and about 12% report owning more than one, the maximum being 7. A little over halfof these are merchant businesses. We also consider the weeks spent in self-employment in theperiod of reference, which is conditional on the household owning an enterprise. On average,households spend about 15 weeks in self-employment - merchant business owner householdsspend the same, and non-merchant business owner households spend 19 weeks.

3.3 Enterprise-level Variables

To analyze the operations of the enterprises which continue to operate when the coffee priceis high, we use several enterprise-level intensive margin variables. These comprise three cat-egories of enterprise operations - capital assets, labor and performance. The first category in-cludes a binary variable for whether the enterprise owns a capital asset and another binaryvariable 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 enterpriseand the number of hired workers. The number of household members active in the businessand the number of hired workers is nearly the same. The performance category consists of three

14

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 continuous measureof profit was a self-reported estimate conditional on the enterprise having a positive amount ofprofit, and so is always positive. The literature on entrepreneurship has highlighted the issueswith accurate measurement of financial flows for micro-enterprises in developing countries (deMel, McKenzie, and Woodruff 2009). We recommend using the same caution in interpreting thecontinuous measure of profits as in the previous literature. The binary variable, which captureswhether positive profits were earned or not, is arguably a more reliable measure of profitability,although it is not measured at as fine a level as the continuous measure.

4 Empirical Strategy

The empirical analysis proceeds in six stages. First, we explore the determinants of businessownership in the pooled sample, by regressing a business ownership dummy on a set of house-hold characteristics meant to capture demographics, financial access, and ability. We estimatethe following simple model, in which o is an ownership dummy and h indexes households:

oh = α+ X′hβ + εh. (22)

Second, we seek to determine the extent to which global coffee prices matter for our sampleof coffee farmers. We do this by testing whether global prices are correlated with the farm-gate coffee prices that farmers face, and whether the global price affects quantities of coffeeharvested, coffee revenues, and household expenditures on food and non-food items.8 The fol-lowing model is estimated, for outcome O, price p, and month (θm), year (δy), and household(µh) fixed effects:

Ohmy = α+ βpmy + µh + δy + θm + εhmy. (23)

As described in the previous section, price p varies at the month x year level. Householdssurveyed in the same month of a particular wave will thus face the same (retrospective) coffeeprice; households that happen to have been surveyed in different months of the same wave willface differing prices.

Third, we ask how fluctuations in the coffee price affect business ownership. In the coffeegrower household sample, we regress an ownership dummy on the coffee price, as well asmonth, year, and household fixed effects using model specified in equation 26.

8Since the farm-gate price that households face is likely endogenously determined (for example, bargainingpower 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 relevantfor our time period in Tanzania), fluctuations over time in the international coffee price should generate exogenouschanges in farm-gate prices, and should thus impact agricultural profitability for coffee-growing households.

15

Fourth, we turn to the individual-level data on business ownership to explore labor supplyresponses to coffee price movements in the (conditional) sample of individuals who reportedowning a business. We regress weeks spent in self-employment in the last 6 months on thecoffee price, year and month fixed effects, and individual fixed effects (ζi), where individualsare indexed by i:

limy = α+ βpmy + ζi + δy + θm + εimy. (24)

The model specified in equation 25 is estimated on a sample whose unit of observation isthus individual x wave, but is only estimated for individuals who reported owning a businessin at least one wave.

Fifth, we examine the responses of capital investments, labor, and performance outcomes ofenterprises to coffee price movements. For this analysis, we use an unbalanced, enterprise-levelpanel. The unit of observation is thus enterprise x wave. The specification used is analogous toequation 25, for enteprises indexed by e and enterprise outcome O:

Oemy = α+ βpmy + ηe + δy + θm + εemy. (25)

Finally, sixth, we examine heterogeneous business ownership responses to coffee price move-ments across various household characteristics. The estimate the following interaction specifi-cation, for household characteristics Xh:

ohmy = α+ γ (Xh x pmy) + βpmy + κXh + µh + δy + θm + εhmy. (26)

Standard errors in all regressions are clustered at the level of the enumeration cluster.

5 Results

5.1 Which households own a business?

We begin by exploring the determinants of household enterprise ownership. We estimate pooledregressions using the whole coffee sample (across all waves). We include the following charac-teristics of household heads: gender (male dummy); numeracy and literacy dummy (need bothfor dummy to equal 1); and a dummy for greater than 0 years of grade completion. For thehousehold, we include the following financial characteristics (all dummies): any remittance ac-tivity (sent or received); positive savings; positive debt; and above-median financial stock.9

The coefficient estimates from the regressions are reported in Table 3. In column 1, we seethat numeracy/literacy is strongly positively related to business ownership. Households with

9Please see the data appendix for details on the construction of these and other variables used in analysis.

16

heads that are both numerate and literate are about 12 percentage points (pp) more likely toengage in entrepreneurial activity, from a mean of about 0.38 (i.e., nearly a 1/3 increase abovethe 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 importanceof dimensions of ability in business ownership and success.

Household financial variables are highly predictive of ownership, and interestingly, all vari-ables we include positively predict ownership. Remittance activity increases likelihood of busi-ness ownership by about 8 pp; positive savings by 11 pp; positive debt by 6 pp; and abovemedian financial stock by more than 8 pp. These results are consistent with the literature onfinancial deepening and microenterprise growth: financial constraints may play a major role inlimiting enterprise development in this context.

Finally, perhaps surprisingly, conditional on numeracy/literacy status, education is not sig-nificantly related to ownership: the coefficient is -0.01 and the standard error band is fairly tightaround 0. Households with male heads are slightly more likely to own a business (5 pp), butthis relationship is again statistically significant.

In columns 2 and 3 of Table 1, we break down enterprises into merchant and non-merchantbusinesses, and look at correlations separately across the two categories. One should think ofthe main delineations as size and fixed cost. Merchant businesses are very small, and usuallyinvolve very little capital input or fixed costs of operation (e.g., many merchants are vendorsthat sell fruits, vegetables, and bread on the roadside). Non-merchant businesses are larger andusually involve bigger capital inputs to production (e.g., a construction business or a drug shopwith a “brick and mortar” location).

Literacy/numeracy seems to matter for both types of businesses, but in both cases the coef-ficient estimates are not precise. Education conditional on literacy/numeracy status continuesnot to play a role in determining ownership. Interestingly, households with male heads aremuch more likely to own non-merchant businesses, but no such relationship exists for merchantbusinesses. 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 op-erate businesses in all four waves of the panel. We estimate this relationship using householdcharacteristics from wave 1, only using household-level observations from wave 1. The depen-dent variable is a dummy indicating “stayer” household. The results here are consistent withthose 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 onliteracy/numeracy is negatively related to “stayer” status, suggesting that perhaps householdswith high innate ability but low grade completion are the most likely to be entrepreneurs in thiscontext.

17

5.2 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 ashousehold 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, im-puted from our transactions data. The results, reported in column 1 of Table 4, demonstrate thatthe 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.10

Next, we examine quantity sold and revenues for coffee. These results, reported in columns2 and 3, show that both quantity and revenue increase with the global price (by about 30 and48 percent, respectively, for a one SD change in the global price), consistent with an upward-sloping supply of coffee.

Finally, we look at household expenditures, to test the extent to which households are ableto smooth consumption in the face of coffee price fluctuations. We find strong evidence againstconsumption smoothing. Total expenditures decline by nearly 20 percent in the coffee farmersample for a one SD drop in the coffee price (column 4). We then split expenditures into foodand non-food categories, and find substantial changes in expenditures in both correspondingto 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).

5.3 Household Enterprise Activity and Coffee Price Fluctuations

Next, we examine whether global coffee price changes affect the probability of household busi-ness ownership. Results are reported in Table 5. Column 1 shows results of a regression of anownership dummy on the coffee price. We find that a one SD drop in the global price increasesthe probability of enterprise ownership by about 5 pp, an increase of about 13 percent above themean. We interpret this finding as strong evidence of countercyclical household enterprise startsin our sample. On average, households are much more likely to engage in enterprise activityduring 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. Theresults show quite strongly that households are much more likely to “smooth” using merchantbusinesses; ownership of non-merchant businesses, on the other hand, does not vary signifi-cantly with the coffee price, and the coefficient is 4 times smaller than the impact of coffee priceon merchant-type business ownership.

10It bears mentioning that we can only impute the farmgate price for households who had non-zero coffee revenuesin the 6-month window prior to survey.

18

5.4 Labor Supply Effects Conditional on Ownership

Next, we examine the impacts of coffee price fluctuations conditional on business ownership, to ex-plore 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 indi-viduals who owned a business. Our measure of entrepreneurial activity intensity is the numberof 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 conditionalon owning at least one business. In this conditional sample, we find a starkly different resulton the impact of coffee price movements. During coffee price booms, entrepreneurial activity,conditional on staying open, actually intensifies: weeks in self-employment increase by nearly3 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 individualsowning merchant and non-merchant businesses, respectively. The coefficient sizes are roughlythe same, but the impact is more precisely estimated in the sample of merchant business owners.

5.5 Enterprise Outcomes Conditional on Ownership

Next, in the same vein as the labor supply analysis above, we examine households’ inputs toenterprise and enterprise performance. The results are reported in Table 7. All analysis is donewithin 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 transactingwith business assets (columns 1 and 2, respectively). It bears mentioning, however, that theseimpacts are imprecisely estimated, and thus we cannot reject the null of a fairly large effect oneither probability at the conventional levels of confidence.

Labor inputs, in contrast, increase significantly with price conditional on enterprise owner-ship. For a one SD increase in the coffee price, the number of household members helping in thebusiness increases by about 0.2 (column 3) and the number of hired workers increases by morethan 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 in-crease in the coffee price generates a 3 month increase in the longevity (months of operation outof the past 6 months) of businesses (column 5); a doubling of business profits (column 6); and a23 pp increase in the probability of positive profits. Overall, the results show resoundingly thatfor those businesses that remain open during booms, inputs and performance increase substan-tially.

19

5.6 Heterogeneous Effects

Last, we explore heterogeneity in the enterprise ownership response to coffee price movementsacross various household characteristics. Table 8 reports results of these interaction specifica-tions. Recall that the main effect from Table 5 showed that households engaged in more en-trepreneurial activity during coffee price busts, and shut down enterprise during booms. Inter-estingly, we find in Table 8 (column 1) that “high-ability” household heads (those with literacyand numeracy) were less likely to change ownership status with coffee price movements. Theimpact of coffee price changes does not appear to be differential across financial activity indica-tors.

When we break down ownership into merchant v. non-merchant categories (columns 2 and3, respectively), we find that ability regulates the relationship between ownership and the coffeeprice only for merchant businesses. Further, higher financial activity and access magnify theimpact of coffee price on ownership during price busts, suggesting that financial constraintsmay hamper the establishment of even these “intermittent” enterprises.

6 Conclusion

Recent studies exploring the drivers of entrepreneurial entry and growth have produced somepuzzling results. The effects of business skills training on growth of existing enterprises appearsmall and insignificant. Returns to capital among microenterprises appear high in some con-texts, but experimental microfinance interventions have produced inconsistent results on en-trepreneurial entry and growth of existing enterprises. On the other hand, studies consistentlyfind large effects of improved access to credit on household consumption.

We propose that many developing households are subject to income shocks, particularlyin agricultural activities, and that they may use microenterprise as a mechanism for weather-ing these shocks. If, for at least a subset of entrepreneurial households, income-smoothing isthe primary motivation of the microenterprise, the above-mentioned results of microfinanceand business training interventions are less surprising. These households may treat capital in-fusions and microenterprise activity as substitutes for smoothing agricultural productivity orprice shocks. “Intermittent” enterprise owners are less likely to avail themselves of enterprisegrowth opportunities, because their businesses do not represent a persistent primary source ofincome.

We find that a rise in the global coffee price significantly decreases the probability of owninga business. We also find that among households who own businesses, a rise in the global priceof coffee drives household members to allocate more productive time to enterprise activity. Wefind that these effects along the extensive and intensive margins are most significant with re-

20

spect to low skill, low capital merchant businesses. Movements along the intensive margin oflabor inputs are mirrored in both the number of household members helping with the businessand the number of hired workers. Though we do not find significant effects on business cap-ital owned, bought, or sold, business performance responds significantly in terms of length ofoperation and profits.

Finally, we find that households with heads who are both literate and numerate exhibitthe strongest effects on the extensive margin (again particularly in merchant businesses), whilehouseholds with educated heads show weaker effects on business ownership. Effects of interac-tions with baseline financial variables are weaker, but suggest that positive savings and higherfinancial stock make households less likely to be entrepreneurial smoothers (that is, less likelyto 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 activityfalls into at least two distinct types in this context: intermittent, smoothing enterprise and persis-tent enterprise. Smoothing enterprise (entrepreneurship in response to agricultural profitabilityshocks) is concentrated in the merchant sector and among households that are literate and nu-merate, but undereducated and active in informal sharing, but have relatively low savings andfinancial stock. Persistent enterprise, on the other hand, is less likely to be in the low capital, lowskill merchant sector and is concentrated among households who are more educated and havehigher savings and financial stock. Additionally, entrepreneurial stayer households actually in-vest more time (both household and hired labor) into their business when agricultural profits arehigh and their enterprises nearly double in profits and duration of operation.

Our results suggest that perhaps allocating public resources toward improving access tocredit for high ability, growth-oriented entrepreneurs might be a high return endeavor, bothin terms of growth of these enterprises and from a general welfare-enhancing perspective. Ofcourse, identifying this high return subset is both very difficult and very important. To thebest of our knowledge, very few studies have attempted to describe microenterprise owners interms of their ability, risk preferences, motivations, expectations, and the like. de Mel, Mckenzie,and Woodruff (2008) is an exception; the authors describe microenterprise owners from a sam-ple in Sri Lanka with respect to some of these dimensions, but are only able to associate thesecharacteristics with enterprise performance and scale in the cross-section. Our study is able tocategorize entrepreneurial households by their heterogeneous optimized enterprise responsesto shocks over time. Our results suggest that those households who do not shut their businesseswhen returns to farming are high must, through an ability-based selection mechanism, be thehighest type households.

21

References

[1] Adhvaryu, Achyuta and Anant Nyshadham. 2013. “Health, Enterprise, and Labor Com-plementarity.” Working Paper.

[2] Banerjee, Abhijit, Esther Duflo, Rachel Glennerster, and Cynthia Kinnan. 2010. “The mira-cle of microfinance? Evidence from a randomized evaluation.” Working Paper.

[3] Crepon, Bruno, Florencia Devoto, Esther Duflo, and William Pariente. 2011. “Impact ofmicrocredit in rural areas of Morocco: Evidence from a Randomized Evaluation.” WorkingPaper.

[4] de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2008. “Returns to Capital inMicroenterprises: Evidence from a Field Experiment.” Quarterly Journal of Economics, vol.123, no. 4: 1329-1372.

[5] de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2008. “Who are the microen-terprise owners? Evidence from Sri Lanka on Tokman v. de Soto.” International Differencesin Entrepreneurship. Joshua Lerner and Antoinette Schoar (Eds.): pp. 63-87.

[6] de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2009. “Measuring microen-terprise profits: Must we ask how the sausage is made?” Journal of Development Economicsvol. 88, no. 1: 19-31.

[7] de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2012. “Business training andfemale enterprise start-up, growth, and dynamics: experimental evidence from Sri Lanka.”World Bank Policy Research Working Paper no. 6145

[8] Kaboski, Joseph and Robert Townsend. 2012. “The Impact of Credit on Village Economies.”American Economic Journal: Applied Economics. forthcoming

[9] Karlan, Dean and Martin Valdivia. 2011. “Teaching Entrepreneurship: Impact of BusinessTraining on Microfinance Clients and Institutions.” Review of Economics and Statistics, vol.93, no. 2: 510-527.

[10] Karlan, Dean, Ryan Knight, and Christopher Udry. 2013. “Hoping to Win, Expected toLose: Theory and Lessons on Micro Enterprise Development.” Working Paper.

[11] McKenzie, David. 2010. “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: 209-233.

[12] McKenzie, David and Christopher Woodruff. 2012. “What are we learning from businesstraining and entrepreneurship evaluations around the developing world?” working paper.

[13] Nyshadham, Anant. 2013. “Learning about Comparative Advantage in Entrepreneurhsip:Evidence from Thailand.” working paper.

[14] Schoar, Antoinette. 2009. “The Divide between Subsistence and Transformational En-trepreneurship.” Innovation Policy and the Economy, Vol. 10, No. 1: pp. 57-81.

22

A Construction of Variables

The following list describes the construction of variables used in analysis. Please note that sincethe survey interviewed households about their decisions and outcomes in the 12 months pre-ceding 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 definedfor these respective time-periods. Furthermore, with the exception of the regression containingthe variable ”Coffee Grower” as the dependent variable, all other regressions are conditional onthe 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, andfor the remaining waves, it is the mean price of the six months preceding the householdinterview. Since different households were interview at different times, it varies acrosshouseholds, though households interviewed in the same month and year will have identi-cal values of robustalag. This is done to align the outcome variables, which were asked forthe previous twelve months in wave 1 and previous six months in the other waves, withthe mean coffee price prevailing in the time-interval when the outcomes were determined.The monthly-level international robusta prices were obtained from the International Cof-fee Organization(ICO).

• Price/SD(Price): a primary independent variable. It equals robustalag divided by thestandard deviation of robustalag. Thus, the coefficients on this variable in the regressionsindicate 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 leastone enterprise, 0 otherwise.

• 1(Household Owns a Merchant Business): equals 1 if the household reported owning atleast 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 owningan at least enterprise of the categories that require semi-skilled and skilled work. Thesecomprise stall keeping, shopkeeping, restaurant owner, garage owner, bus driver, black-smith, plumber, carpenter, tailor, repair work, mechanic, mason, painter, hair dresser,shoemaker, butcher, handicrafts, photographer, and doctor.

23

• Stayer: equals 1 if the household reported owning an enterprise in all 4 waves, 0 otherwise.

A.3 Characteristics of the Head of the Household

• 1(Male): equals 1 if the head of the household is male, 0 otherwise.

• 1(Can Write and Do Math): equals 1 if the household head reported being able to write aletter and perform written calculations, zero otherwise.

• 1(Some Education): equals 1 if the household head reported attending primary schoolingor above for any length of time, including adult or Koranic education, 0 otherwise.

A.4 Household’s Financial Participation Variables

• 1(Any Remittances): equals 1 if the household reported either sending or receiving remit-tances.

• 1(Positive Savings):equals 1 if the household reported having saving greater than zero, 0otherwise.

• 1(Positive Debt): equals 1 if the household reported having debt greater than zero, 0 oth-erwise.

• 1(Above Median Financial Stock): equals 1 if level of financial stocks (Savins + loan debt)reported by the household are greater than the median level of financial stocks over theentire sample, which is 2,000 TZS.

A.5 Household variables related to coffee cultivation

• Coffee Grower: equals 1 if the household reported harvesting coffee in a particular wave,0 otherwise.

• ln(1+ Harvest Area Under Coffee): natural log of one plus the number of acres the house-hold reported harvesting under coffee.

• ln(Price Received by Households): natural log of value of coffee sales reported by thehousehold divided by the quantity sold in USD/kg (using the exchange rate of iTZS=0.0006USD).

• ln(1+Quantity Sold): natural log of 1 plus the quantity of household-level coffee salesreported in kg.

• ln(1+Coffee revenues): natural log of 1 plus the value of household-level coffee sales re-ported.

24

A.6 Household expenditure variables

• ln(1+Total Expenditure): Total household expenditure, inclusive of food and non-foodexpenditure (detailed below), consumption of home production, remittances sent, andimputed expenditure for wage income in-kind.

• ln(1+Food Expenditure): The sum of seasonal and non-seasonal food expenditure, inclu-sive of expenditure on meals and beverages consumed away from home, exclusive of con-sumption of home production.

• ln(1+Non-Food Expenditure): The sum of expenditure on education, health, housing, util-ities, funeral and other non-food expenses.

A.7 Household labor outcome variables

• Weeks in Self-Employment: The number of weeks reported at the individual-level spentin non-farm self employment in the reference period described above, aggregated up tothe household-level .

A.8 Enterprise-level intensive margin variables

• 1(Business Assets Owned): equals 1 if the enterprise owns at least one of the followingcategory of assets: a) buildings and land, b) vehicles or boats, c) tools, equipment or ma-chinery, or d) other durable assets for use in the enterprise, 0 otherwise.

• 1(Business Assets Bought or Sold): equals 1 if any asset described above was bought forthe enterprise or divested from the enterprise.

• Number of Household Members Helping in the Business: The number of household-members who helped in the enterprise, including those were unpaid.

• Number of Hired Workers: The number of hired workers working in the enterprise.

• Mos. Business Operating: The number of months in the reference period described abovethat the business was operating.

• ln(1+ Business Profits): The natural log of one plus business profits. Business receipts,such as revenues and profits, were asked for the last 12 months in wave 1 and the last 6months in waves 2, 3 and 4. In case the business had begun to be operational after this timeframe, the usual receipts per day, week or month were asked. Hence, the profits variablewe use equals the former measure if the enterprise had been in operation in the respective

25

time-period, and the latter measure if enteprise operations began after the associated timeperiod. We convert all profits to the daily level.

• 1(Business Had Positive Profits): equals 1 if the enterprise had profits greater than zero, 0otherwise. Profits are constructed as defined above.

26

Number of household‐year observationsNumber of households

Mean SD Mean SD Mean SDEnterprise Activity 1(Household has a merchant business in specified period) 0.244 0.429 0.508 0.500 0.460 0.499 Months business has been operating 4.110 2.713 4.817 2.822 1(Business Assets Owned) 0.650 0.477 0.766 0.424 1(Business Assets Bought or Sold) 0.146 0.353 0.172 0.378 Household members helping with business 0.463 0.840 0.457 0.800 Hired workers 0.408 1.574 0.516 1.487 1(Business had positive profit) 0.465 0.499 0.540 0.499 Business Profits 106.450 483.630 125.488 385.215

Household Head Characteristics 1(Male) 0.735 0.441 0.804 0.397 0.837 0.370 1(Can write AND do math) 0.708 0.455 0.799 0.401 0.823 0.382 1(Some Education) 0.813 0.390 0.894 0.308 0.880 0.325

Financial Resources 1(Remittances Received) 0.857 0.350 0.894 0.308 0.899 0.301 1(Remittances Sent) 0.975 0.155 0.990 0.100 0.992 0.090 1(Positive Savings) 0.851 0.356 0.935 0.247 0.943 0.232 1(Positive Debt) 0.465 0.499 0.540 0.499 0.554 0.498 1(Above median financial stock) 0.507 0.500 0.641 0.480 0.677 0.468 1(Above median physical stock) 0.554 0.497 0.629 0.483 0.655 0.476

Notes: Except for variables on Enterprise Activity which are at the enterprise level, all other variables are at the household level.

Table 1Summary Statistics: Enterprise Activity, Demographic Characteristics, and Financial Resources

Households with a business in at least 1 period

1,087

Whole Panel SampleHouseholds with a business in

all 4 waves (Stayers)

2,880 368753 465 92

1(Enterprise Ownership)

Enterprise Histories 1(Household has a business in wave 1) 1(Household has a business in wave 2) 1(Household has a business in wave 3) 1(Household has a business in wave 4) 1(Household ever switched enterprise status)

Proportion of Households with Enterprises in: 0 waves 1 waves 2 waves 3 waves 4 waves

Coffee PriceInternational Robusta Coffee PriceInternational Robusta Coffee Price in 1990International Robusta Coffee Price in 1991International Robusta Coffee Price in 1992International Robusta Coffee Price in 1993

Notes:

42.658 4.4517.33552.497

53.603 2.76

Whole Panel Sample

49.344 6.309

48.621 2.3

Mean SD

0.5000.473

0.723 0.448

0.122 0.328

0.04 0.1960.043 0.2020.073 0.260

Table 2Summary Statistics: Enterprise Histories and Coffee Price

Whole Panel Sample

Mean

0.2780.3710.4390.432

SD

0.448

0.377 0.485

0.4830.4970.496

Panel A: Household Head Characteristics

1(Household Owns a Business)

1(Household Owns a Merchant Business)

1(Household Owns a Non‐Merchant Business)

Stayer: 1(Business in all 4 Waves)

1(Male) 0.0476 ‐0.00731 0.0655*** 0.0402(0.0334) (0.0293) (0.0232) (0.0274)

1(Can Write and Do Math) 0.117** 0.0548 0.0655* 0.0890***(0.0443) (0.0337) (0.0334) (0.0273)

1(Some Education) ‐0.0133 0.0133 0.00726 ‐0.0756**(0.0369) (0.0371) (0.0304) (0.0289)

1(Any Remittances) 0.0799** 0.0189 0.0761*** 0.0392(0.0341) (0.0305) (0.0225) (0.0255)

1(Positive Savings) 0.113*** 0.0412 0.0740*** 0.0609**(0.0272) (0.0250) (0.0213) (0.0256)

1(Positive Debt) 0.0556** 0.0582** 0.0145 0.0321*(0.0264) (0.0275) (0.0182) (0.0186)

1(Above Median Financial Stock) 0.0842*** 0.0819*** 0.0345 0.0490**(0.0274) (0.0199) (0.0254) (0.0240)

Observations 2,876 2,876 2,876 752Mean of Dependent Variable 0.377 0.237 0.217 0.122

Table 3Which Households Own a Business?

Notes: Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Household characteristics are from wave 1.

ln(1+Food Expenditure)

ln(1+Non Food Expenditure)

Price/SD(Price) 0.240** 0.293** 0.478** 0.188*** 0.219*** 0.151*** ‐0.00573 0.00856(0.113) (0.116) (0.192) (0.0313) (0.034) (0.0303) (0.00543) (0.0131)

Fixed Effects

Observations 668 1,795 2,879 2,837 2,837 2,837 3,515 2,879Number of Households 486 714 753 753 753 753 975 753Mean of Dependent

Variable175.3 63.52 25,069.49 244,716.90 44,211.48 94,148.61 0.761 0.605

Table 4Effects of Coffee Price Changes on Coffee Production and Household Expenditures

Household, Year & Month

Notes: Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Coffee grower is a dummy variable that is 1 if a household reported harvesting coffee in that wave. Coffee sample is a dummy variable that is 1 if a household reported harvesting coffee in any one of the 4 waves, and 0 otherwise. Harvest area under coffee is the number of acres harvested in the last 12 months if the household is surveyed in the first wave, and the number of acres harvested in the last 6 months if the household is surveyed in any subsequent wave. The mean of all dependent variables are reported in levels e.g. in the column for ln(food expenditure), the mean of the food expenditures in Tanzanian shillings is reported.

ln(Price Received by Households)

ln(1+Quantity Sold)

ln(1+Coffee Revenues)

ln(1+ Harvest Area Under Coffee)

Household, Year , & Month

ln(1+Total Expenditure)

Total Expenditure Coffee Grower

Price/SD(Price) ‐0.0511*** ‐0.0428*** ‐0.0148(0.00991) (0.0118) (0.0111)

Fixed Effects

Observations 2,879 2,879 2,879Number of Individuals 753 753 753Mean of Dependent

Variable0.377 0.237 0.217

Does Household Enterprise Activity Respond to Coffee Price Fluctuations?Table 5

Household , Year and Month

Notes: Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Coffee grower is a dummy variable that is 1 if a household d h ff h ff l d bl h f h h ld d h ff f

1(Houshold Owns a Business)

1(Houshold Owns a Non‐Merchant Business)

1(Houshold Owns a Merchant Business)

Price/SD(Price) 2.975** 3.587** 3.084(1.301) (1.691) (1.838)

Fixed Effects

Observations 1,087 683 626Number of Individuals 465 354 339Mean of Dependent

Variable 15.91 15.37 19.09

Notes: Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Weeks in self‐employment in the last 12months for the first wave and the last 6 months in all subsequent waves.

Table 6Does Household Enterprise Activity Respond to Coffee Price Fluctuations?

Conditional on Owning a Business

Conditional on Owning a Merchant Business

Conditional on Owning a Non‐Merchant Business

Weeks in Self‐Employment

1(Business Assets Owned)

1(Business Assets Bought or Sold)

Number of Household Members Helping in

the Business

Number of Hired Workers

Mos. Business Operating

ln(1+ Business Profits)

1(Business Had Positive Profits)

Price/SD(Price) 0.0884 0.0589 0.192* 0.419** 3.096*** 0.984** 0.233**(0.0890) (0.0554) (0.107) (0.160) (0.666) (0.381) (0.0995)

Fixed Effects

Observations 1,595 1,594 1,595 1,595 1,595 1,595 1,595Mean of Dependent Variable 0.650 0.146 0.463 0.408 4.108 137.2 0.465

Notes: Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). The mean of all dependent variables are reported in levels e.g. in the column for ln(Business Profits) the mean of the profits in Tanzanian shillings is reported.

Do Enterprise Inputs and Performance Respond to Coffee Price Fluctuations?Table 7

Capital PerformanceLabor

Household, Year and Month

1(Male) x Price/SD(Price) ‐0.0190 0.00953 ‐0.0230(0.0152) (0.0170) (0.0158)

1(Can write AND do math) x Price/SD(Price) 0.0411** 0.0573*** ‐0.00640(0.0197) (0.0177) (0.0182)

1(Some Education) x Price/SD(Price) ‐0.0129 ‐0.0363** 0.0242(0.0166) (0.0160) (0.0168)

1(Any Remittances) x Price/SD(Price) ‐0.00485 0.00196 ‐0.0125(0.0246) (0.0216) (0.0166)

1(Positive Savings) x Price/SD(Price) ‐0.0142 ‐0.0309* 0.0245*(0.0160) (0.0168) (0.0143)

1(Positive Debt) x Price/SD(Price) 0.0211 0.000842 0.00910(0.0138) (0.0146) (0.0102)

1(Above median financial stock) x Price/SD(Price) ‐0.0198 ‐0.00806 ‐0.0226*(0.0131) (0.0128) (0.0130)

Price/SD(Price) ‐0.0394 ‐0.0333 ‐0.0153(0.0286) (0.0277) (0.0232)

Fixed Effects Household, Wave, & Month Household, Wave, & Month Household, Wave, & Month

Observations 2,875 2,875 2,875Number of Households 752 752 752

Mean of Dependent Variable 0.377 0.237 0.217

Notes: Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Coffee grower is a dummy variable that is 1 if a household reported harvesting coffee in that wave. Coffee sample is a dummy variable that is 1 if a household reported harvesting coffee in any one of the 4 waves, and 0 otherwise. Household characteristics are from wave 1.

Are Household Enterprise Activity Responses Heterogeneous by Household Characteristics?Table 8

1(Houshold Owns a Business)1(Houshold Owns a Merchant

Business)1(Houshold Owns a Non‐

Merchant Business)