«Abstract We explore which ﬁnancial constraints matter the most in the choice of becoming an entrepreneur. We consider a randomly assigned welfare ...»
We mainly concentrate on the ﬂows into entrepreneurship, i.e. on those individuals who are either salaried or unemployed at the baseline and become entrepreneurs in the follow-up period. Amongst those residing in control villages, 4% become entrepreneur during in this period, of which roughly 25% were unemployed in the baseline and 20% are women. Agricultural assets are their main capital endowment: 54% of those who become entrepreneurs own productive land, 25% own working animals such as horses, donkeys and bullocks. These ﬁgures are however very similar to the asset holdings of salaried workers.
A distinctive features of new entrepreneurs is instead their engagement in micro-business activities not (directly) related to agriculture: 11% of new entrepreneurs declare to be engaged in activities like handicraft, sewing clothes and domestic services, whereas the corresponding share for salaried workers is only 3%.
Moreover, we note that 34% of new entrepreneurs have more than one paid occupation vis-`-vis 8% of salaried workers. This is common in many developing settings, and it is a typically interpreted as an income smoothing strategy (see e.g. Morduch , Banerjee and Duﬂo ). Indeed, also in our sample, new entrepreneurs face a substantially higher volatility of labor income in their primary occupation, which may increase their need for self-insurance.3 3 The standard deviation of monthly labor income is 84% of the sample mean for entrepreneurs vis-`a vis 60% for salaried workers.
6 3 Entrepreneurship and Financial Constraints Random treatment assignment implies that a simple comparison of treated-control mean outcomes will likely provide an unbiased estimate of the program impacts. However, we additionally control for several socioeconomic characteristics that may aﬀect occupational choices so as to improve the power of the estimates and check the robustness of our ﬁndings. Moreover, although villages were randomly assigned to the treatment, data are unlikely to be independent across individual observations. In particular, occupational choices of individuals in the same village may be correlated as they share background characteristics and are exposed to the same market environment and natural shocks. In this section, we ﬁrst introduce a standard reduced-form empirical framework to evaluate whether the exposure to the treatment induces some individuals to become entrepreneur.
We then provide some additional evidence to suggest that the program impacts arise from individual responses to the cash transfers rather than to the conditions (such as schooling behaviors) attached to these transfers.
3.1 Treatment Impacts Consider an individual i who is either salaried worker or unemployed in the baseline and let ne∗ be a dummy equal to one if the individual has become entrepreneur in a i,t given post-program wave t and zero otherwise. Suppose ne∗ is determined by the latent i,t variable nei,t, which denotes individual i’s probability of becoming entrepreneur. We then
estimate regressions of the following form:
nei,t = αTl + Xi,t0 γ + δt + ηs + i,t, (1)
where Tl represents the Progresa experimental treatment assignment at the locality level l and the vector Xi,t0 denotes a set of pre-determined covariates at the individual, household and locality level: individual age, gender, education, income, spouse main occupation, household wealth and demographic composition, village shares of entrepreneurs and proxies for agricultural risk. We also include wave dummies and state dummies, δt and ηs.4 In order to take into account the potential intra-village correlation of mentioned i,t above, we cluster standard errors at the village level.
Table 2 reports probit marginal eﬀects of the program on the transition into entrepreneurship. Treatment impacts appear to be both statistically and economically signiﬁcant. As shown in columns (1) and (2), living in a treated community increases the probability of entering self-employment by 0.7 percentage points. This represents 4 We cannot specify ﬁxed eﬀects at a more disaggregated geographical level -say, municipality or village- since this would imply loosing the exogenous variation induced by the experiment.
7 an increase of 19% with respect to the counterfactual sample averages (equal to 4%).
In columns (3)-(6), we show that the treatment signiﬁcantly increases the probability of entry into entrepreneurship both from salaried work and from unemployment. In relative terms, the eﬀects across subsamples are of comparable magnitudes: having access to this stable source of extra income increases the likelihood to become entrepreneur of about 20% in the program period.
As further evidence that the above results are due to the treatment, we also include the period in which control villages are incorporated into the program (survey waves 4 and 5), and we slightly modify equation (1) so as to allow for interaction eﬀects of the treatment indicator with each program wave dummy. The results provided in Table 3 (columns 1-2) show that indeed treated-control diﬀerences tend to vanish once the control group is incorporated. We also investigate whether our results may be driven by a pure demand eﬀect, whereby treated villages are richer and so have an higher demand for entrepreneurs. If this were the case, the treatment eﬀect would hold for all households in a treated village, whether eligible for the program or not. The results provided in columns 3-4 do not support this hypothesis: there are no treated-control diﬀerences for non-eligible individuals. It appears that being entitled to receive the treatment, as opposed to simply living in a treated village, is what drives our eﬀects. This result also tends to exclude within-village spillovers between eligible and non-eligible households in the choice to become entrepreneur.
3.2 Conditionality As described in Section 2, cash transfers are conditional on health and schooling behaviors. In particular, the requirement of sending children to school may have a direct eﬀect on occupational choices: for example, as children are less likely to work at home, mothers may have to quit a salaried job and turn to self-employment in search for ﬂexible working hours or home working. This seems however unlikely to drive our results. First, among those who were salaried and became entrepreneurs in treated villages, only 6% have children who returned to school in the post-program period, of which only 5% are women.
Moreover, as shown in Table 4, we ﬁnd no diﬀerential program impacts on individuals for whom, according to a series of pre-program characteristics, we expect conditionality to be more or less binding.5 We also notice that if individuals were pushed into entrepreneurship because of conSpeciﬁcally, we consider those who were working longer hours, those who had eligible children not enrolled in school (who had to actually change their behavior in order to receive the treatment), those who had eligible children only in primary school age vs. those who had female children in secondary school age (enrollment in primary school is very high irrespective of the treatment, while the treatment has a bigger eﬀect on female secondary schooling; see Schultz ).
8 ditionality, we would expect their labor supply to change and in general their welfare to decrease. The results presented in Table 5, however, oﬀer little support to this hypothesis. Despite these results should be interpreted as simple correlations, as the choice of becoming entrepreneur is in itself dependent on the treatment, they show that new entrepreneurs in treated villages have signiﬁcantly higher labor earnings and higher nonfood expenditures (columns 1-2), not signiﬁcantly diﬀerent food consumption and labor supply (columns 3-5), and they are less likely to be engaged in a second paid occupation (column 6).6 Taken together, this evidence tends to rule out that conditionality is driving our results, so we can interpret the program impacts as the result of an income shock and thus as (indirect) evidence that households face ﬁnancial constraints.
4 Liquidity and Insurance Constraints Absent the program, individuals may refrain from becoming entrepreneurs for at least two reasons. First, they may face liquidity constraints which prevent them from undertaking some initial capital investment. The program would then promote entrepreneurship by increasing households’ current liquidity. Second, individuals may prefer avoiding the risk associated with entrepreneurial returns. In this case, by providing transfers for an extended and predictable period of time, the program would promote entrepreneurship by increasing households’ ability to cope with future income ﬂuctuations. In this section, we ﬁrst develop a simple model to highlight how liquidity and insurance constraints respond diﬀerently to the time proﬁle of expected income shocks. We show that, under standard assumptions, the choice of becoming entrepreneur is more responsive to the amount of transfers currently received if liquidity constraints are binding, while it is more responsive to the amount of transfers expected for the future if insurance constraints are binding.
We then empirically explore these mechanisms by taking advantage of a second source of variation. As described in Section 2, beyond random treatment assignment, households diﬀer in the magnitude and time proﬁle of the transfers they are entitled to, as determined by the number, grade and gender of their children. We can then test whether new entrepreneurs are more responsive to the amount of money currently received or to those expected for the near future.
These results come from estimating, for each of output yi,t, the parameter γ in the following equation:
yi,t = αTl + βnei,t + γTl ∗ nei,t + Xi,t0 λ + ui,t.
A similar strategy is used to get a sense of how new entrepreneurs invest the money. We notice in Table 6 that there is no evidence of increased investment in agricultural activities, such as acquisition of land, animals or agricultural expenditures or production. On the other hand, there is evidence of increased nonagricultural activities, in particular carpentry and handicraft.
4.1 A Simple Occupational Choice Model Consider a population of individuals who are heterogeneous in their initial wealth a and in their risk aversion r, drawn respectively by smooth distributions F and G with density
f and g. Individuals live for two periods. In the ﬁrst period, they choose their occupation:
either they become self-employed, which requires a ﬁxed investment of k units of capital, or they become salaried. In addition, they choose the amount of wealth they wish to save from period 1 to period 2. We allow savings to be positive or negative and normalize the net returns of both saving and borrowing to zero. We denote with se the amount of savings decided by an individual in case he becomes entrepreneur and with sw the amount decided in case he becomes a worker.
In the second period, individuals enjoy the returns from their occupation. The selfemployed get y with probability p and zero otherwise; salaried workers get a ﬁxed wage w, where py − k w.
Savings and occupation are chosen in order to maximize
where x1 and x2 denote consumption in period 1 and 2. We make the standard assumption that u exhibits decreasing absolute risk aversion (DARA) and for simplicity we abstract from time discounting. Finally, irrespective of their choices, individuals are entitled to cash transfers C1 in period 1 and C2 in period 2.
An individual becomes entrepreneur if his expected utility exceeds what he would enjoy as a worker, where this diﬀerence writes E = u(a−k −se +C1 )+pu(se +y +C2 )+(1−p)u(se +C2 )−u(a−sw +C1 )−u(sw +w+C2 ).
4.1.1 Equivalence between current and future transfers Suppose there are no constraints on borrowing or saving. Individuals who become workers set sw such that their marginal utility is equalized across periods, i.e.
and so occupational choices respond in the same way to current and future transfers.
This result is not surprising. In a world in which wealth can be freely and costlessly allocated across periods, individuals see no fundamental diﬀerence between the transfers they have received today and those they known they will receive tomorrow.
We do not expect however this to be typically the case. Borrowing constraints are widely documented (restricting to developing countries, see the surveys in Banerjee  and Karlan and Morduch ). Households may also face saving constraints, as the result of present-biased preferences (Ashraf, Karlan and Yin , Dupas and Robinson , Banerjee and Mullainathan ), social norms (Platteau ), or simply unavailability of a safe storage technology (see Collins et al.  and the survey by Karlan and Morduch ). We then turn to a setting in which some individuals may face constraints in their choice of se and sw. These constraints break the equivalence between current and future transfers and allow us to compare the eﬀects of these transfers in two extreme cases: one in which there are only liquidity constraints (k 0 and individuals are risk neutral) and one in which there are only insurance constraints (k = 0 and individuals are risk averse).
4.1.2 Liquidity constraints Consider ﬁrst the case in which individuals are risk neutral. Given that in this case everyone would like to become entrepreneur and so invest in period 1, saving constraints
4.1.3 Insurance constraints We now
from liquidity constraints by assuming k = 0. In this case, all those who are suﬃciently tolerant toward risk become self-employed, i.e. ne = G(r∗ ). Suppose there are (extreme) borrowing constraints (se ≥ 0 and sw ≥ 0) so that some individuals, even by not saving, are consuming too little in the ﬁrst period (as they would like to borrow). This requires that for sw = 0
where ∂r∗ /∂C2 0 follows from the fact that u is DARA and so increasing C2 increases risk-taking through a classic wealth eﬀect (Pratt ). Those for whom only (8) holds set se 0 and sw = 0 and they too are more responsive to future than to current transfers.