Give Me a Pass: Flexible Credit for Entrepreneurs in Colombia

Although microcredit has reached millions, recent randomized evaluations find limited average business impacts. Contract rigidity, specifically the fixed and frequent installments, may limit productive risk-taking and thus diminish impact on average profits but risk triggering moral hazard. We test this with a Colombian lender that experimentally compared, for a sample of new borrowers, rigid lending to a loan product that included three “passes” to push off a monthly payment to the future. The flexible loan did lead to some shifts in investment behavior but no average impact on revenue or profits level or variance, and did lead to higher default.


Introduction
Most small firms in developing countries have large month-to-month fluctuations in their income stream and thus cash flow.Anticipated cash shortfalls due to seasonality, as well as unanticipated positive and negative shocks such as business opportunities, health shocks, etc., contribute to this volatility.
In credit markets with full information, lenders would "match cash flows", i.e., provide credit terms that tailor disbursals and repayments to a firm's cash fluctuations.A working capital line of credit is a simple example.More complex structures in this spirit are offered by venture capitalists or revenue-sharing contracts with repayments linked to firm performance (Gompers and Lerner 2001).In credit markets with information asymmetries, such as those in developing countries, lenders still try to match repayment to cash-flows accounting for seasonality or observable shocks.For example, most agricultural loans are offered with a single installment due at harvest as farmers typically receive income only after the crops are sold.Idiosyncratic, unanticipated shocks, however, are harder to verify; perhaps because entrepreneurs could misreport actual revenues, full revenue-sharing contracts seem nonexistent (de Mel,  McKenzie, and Woodruff 2019; Cordaro et al. 2022).
Many microentrepreneurs seeking formal credit in developing countries rely on microcredit loans with fixed, frequent repayments that start immediately after the loan is disbursed (Armendariz de Aghion and Morduch 2010; Labie, Laureti, and Szafarz 2017).Borrowers may adjust to these terms by holding cash back or by passing on high (risk-adjusted) return investments (Karlan and Mullainathan 2007; Field et al.  2013; Fischer 2013).And, perhaps due to this rigidity, microcredit loans have had limited impacts on the profitability and growth of firms (Banerjee, Karlan, and Zinman 2015; Crépon et al. 2015) although impacts at scale for the full industry (versus marginal shifts by one lender) have been shown to generate larger impacts (Breza and Kinnan 2021).
Recent attempts to introduce repayment flexibility to existing clients have shown that flexibility can improve business outcomes without deteriorating repayment rates (Battaglia, Gulesci, and Madestam  2023; Barboni and Agarwal 2023).This may not be true for first-time borrowers: providing flexibility could backfire for the lender if some initial fixed and frequent repayment loans are needed to screen or teach discipline in repayment.On the other hand, flexibility could attract new, (in expectation) profitable clients uninterested in the standard microcredit loan due to its rigidity.Indeed, those rejecting rigidity may reveal a high personal cost of default (e.g., due to personal ethics or reputation) and such clients are quite desirable for the bank.If the share of such entrepreneurs is large, flexibility should be offered to new borrowers.We thus seek to assess the validity of these theories on new borrowers by evaluating experimentally the impact of repayment flexibility on selection, client welfare, and loan performance.
We collaborate with a microlender in urban Colombia to introduce repayment flexibility in a two-stage offer-contract design to new clients.The flexible credit feature allows borrowers to use a "pass" at any time during the loan, allowing them to only pay the interest amount of an installment, postponing the payment of the principal amount, up to three times on a 12-month loan.The experimental design employs three treatment arms: (1) Flex→Flex is offered and disbursed the flexible credit, (2) Standard→Flex is offered the standard credit but then surprised with the flexible credit at disbursement, and (3) Standard→Standard is offered and disbursed the standard rigid credit.This allows us to test both for selection effects as well as contract effects on choices and outcomes after borrowing.
We report three main findings.First, there are no selection differences in the take-up rates, characteristics, or outcomes of the Flex→Flex group compared to the Standard→Flex group.The lack of selection effects suggests only a small share of profitable entrepreneurs would reject the standard contract but accept the flexible contract.Second, flexibility increases default ---and the effect is driven by borrowers who used the flexibility to extend loan maturity and had already missed payments at the time of default.Comparable borrowers in the control group had better repayment performance without resorting to more expensive sources (i.e.informal loans).Third, flexibility leads to more self-reported client satisfaction but not to higher retention of successful borrowers.
We contribute to the small but growing literature investigating flexibility in microcredit contracts (see Appendix Table 1 for a summary of the features of five similar studies).Field et al. (2013) finds that providing borrowers with an initial two-month grace period leads to higher-return (and higher-risk) investments.While the grace period leads to higher long-run profits for the borrower, it is not profitable to the lender, which suffers the downside of the increased risk without the upside benefit of increased returns.This mechanism of higher-risk investments leading to increased defaults likely does not apply in our setting as we find no difference in the variance of sales or profits between borrowers of flexible and standard loans.In addition, clients in Field et al. (2013) could not choose whether or not to use the grace period, and thus the paper is less applicable to assess the impact of flexibility per se on borrower discipline and repayment norms.Barboni and Agarwal (2023) shows that a three-month block repayment holiday, communicated in advance and available upon successful repayment of three monthly installments of a 24 months loan, attracts financially disciplined clients and leads to higher sales and repayment rates.Since the intended use of the repayment delay had to be communicated to the microcredit institution by the borrower one month in advance, the product flexibility only targets anticipated income fluctuations or profit opportunities.The flexible loan product that is closest to ours is studied in Battaglia et al. (2023).Borrowers who were deemed eligible by loan officers based on their repayment histories were given two passes (versus three in our setting) on a 12-month loan that could be used at any point during the loan tenure, catering to both unexpected shocks and predicted downturns.The flexibility led to improvements in business and socioeconomic status and lower default rates, especially for borrowers operating smaller businesses.A critical difference between the literature cited above and our study is whether the sample were new or experienced clients: we study new clients in order thus allowing us to better study both the selection effect explicitly as well as a population that has not yet demonstrated financial discipline, whereas the above literature studies current (and previously successful with respect to repayment) borrowers at each of the partner lenders.Field et al. (2013) is the exception and studies both new and existing borrowers. 1 showing no evidence of selection effects from introducing flexibility to new clients, we also contribute to the literature assessing the extent of selection in low-income country credit markets (see e.g., Karlan  and Zinman 2009; Ahlin et al. 2020; Beaman et al. 2020; Gertler, Green, and Wolfram 2023; Jack et al.  2023).

Setting and the Standard Credit Product
We partnered with the microcredit unit of Fundación Mario Santo Domingo ("FMSD"), a small not-forprofit lender.FMSD operates in northern Colombia and had around 6,000 clients.The experiment took place in the urban branches of Barranquilla and Cartagena.FMSD gave individual liability loans to both male and female entrepreneurs for either working capital or the purchase of business fixed assets.Eligible borrowers had to own an existing business for at least six months, had to be in good standing with the credit bureau, and could have at most one other loan with another institution.Loans given by FMSD required fixed monthly installments and had no early repayment penalties.The median and modal loan length was 12 months but varied from six to 24 months.The nominal interest rate ranged from 36% p.a. to 42% (see Appendix A: Details of Experiment for details) plus various fees amounting to 14% of the principal for a typical loan (see Appendix Table 2 for details).Borrowers with a past due balance at the end of the month lost access to a lower interest rate reserved for successful repeat borrowers and were reported to the credit bureau.Borrowers with two or more months with a past due balance were denied future loans.

The Flexible Credit Product
In collaboration with the lender, we developed a new credit product with repayment flexibility.Specifically, the flexible credit introduced "passes" that allowed borrowers to pay only the interest and fees of the monthly installment, postponing the principal portion without penalties for missed payments.The delayed principal amount accrued interest at the same rate as the original loan and was subsequently due either at the end of the loan (thus extending the term) or earlier, as the borrower chose.2Borrowers were allocated one pass for every four months of the initial loan duration.A borrower with the typical 12-month loan, for example, would be given three passes that could be used at any point in the loan cycle, including sequentially.To use a pass, borrowers had to contact their credit officer via phone or in person by visiting the branch before the payment was due that month.
Each time a borrower used a pass, he or she chose between two different types depending on how the principal was repaid.If the client used an "extension" pass, the loan maturity was extended by one month without changing the amount of the remaining monthly installments.Alternatively, under the "noextension" pass, clients paid the postponed principal (plus accruing interest) in one or more payments within the original loan term.Appendix Table 2shows example repayment schedules for extension and no-extension type passes.Given that the installment amount was fixed during the repayment schedule, the share of installment due to the principal payment increased over time and so did the amount that was skipped with the pass.
Except for the repayment flexibility, the new credit product was identical to the standard credit offered by the lender.

Experimental Design
Figure 1 provides an overview of the two-stage experimental design.In the first stage, potential first-time clients were offered either a standard loan or a flexible loan.All offers were subject to the lender's standard loan approval process.In the second stage, conditional on completing the application and subsequent approval, a share of standard loan clients were switched to a flexible loan by surprise (Karlan  and Zinman 2009).As a result, our design has three experimental groups: "Flex→Flex", "Standard→Flex", and "Standard→Standard".
We chose this two-stage design to disentangle selection effects from contract effects.To study selection effects, we analyze outcomes for borrowers that end up with a flexible contract and compare "Standard→Flex" clients --who received the standard loan offer but were later switched to a flexible loan---with "Flex→Flex" clients who were offered the flexible loan from the beginning.To study contract effects, we analyze outcomes for borrowers offered the standard loan and compare credit outcomes of "Standard→Flex" clients with "Standard→Standard" clients.

Sample recruitment and randomization of first stage (initial offers)
We worked with FMSD to integrate the randomization of initial flexible offers in their recruitment of firsttime clients.In total, 8,610 potential clients were approached for initial offers.Panel A of Appendix Table 3 reports the share of potential clients recruited through the different channels used by the lender.About half of the offers were made by "door-to-door" promoters. 3In total across promoters, credit officers and front desk staff, about 30% of potential clients were recruited during public "financial" events organized by the local mayor's office or directly by FMSD. 4 The remaining pool of potential clients, about 20%, were called up by credit officers directly or visited the branch.
Once potential clients were registered, credit officers followed-up with a visit to assess eligibility, and to make randomized offer.All prospective clients also received a leaflet with information about the loan (see Appendix B for a sample of the flex product flyer in Spanish).Loan applications were collected by credit officers and reviewed by the credit committee.Clients with approved loans received additional explanations from a dedicated staff when the loan was disbursed either during the branch visit or over the phone.
Recruitment into the study took place continuously over 18 months.Overall, 22.4% of potential clients were assigned to a flexible offer (see Appendix A for further details).Panel A of Appendix Table 3 confirms that the randomized assignment of offer types was balanced overall with respect to the recruitment process and branch location (the p-value of a joint test of equality of means is 0.23).

Randomization of second stage (switch to flexible loans)
Approved standard loans were randomly switched to flexible loans at disbursement, with a target probability of 50%, based on the observed distribution of the last three digits of the national identification document using the loan data set of our partner microcredit institution.In total, 1,893 standard loan offers were accepted and 971 (51%) of them were converted to flexible contracts as part of the second stage randomization.
Clients learned about the switch when their credit officer called them about the approval of their application and gave a short explanation of the new flexible loan.All clients in the Std→Flex group accepted the switch to the flexible loan.
We test for balance in the second stage randomization by looking at the sample of new clients that initially received a standard offer.Using a combination of data from the recruitment process, data collected by credit officers during the application process as well as the bank's administrative data, we compare those who received a standard loan with those who were switched to a flexible loan.Appendix Table 4 shows means and standard deviations for the two groups and p-values of the tests of equal means.Out of the 18 variables including loan characteristics (Panel A), socioeconomic characteristics of clients (Panel B) and business characteristics (Panel C), only one difference is significant at the 10% level.The p-value of a joint test of differences across all variables is 0.90.

Data
We draw on several data sources.First, we use self-reported data (on household and business characteristics) collected by credit officers at the time of the loan application.Second, we use administrative data with loan characteristics and client repayment histories for all study loans.The data cover 100% of clients from loan disbursement until three months past loan maturity (and 99.3% until 12 months past maturity), with loan maturity accounting for extensions due to passes.The data span 49 months from when the first loans were disbursed until 30 months after the last set of loans were disbursed.
Third, the lender conducted client satisfaction phone surveys on a subsample of study clients.The lender's staff called both standard and flexible loan clients to assess client attitudes towards their loan product, their level of knowledge about the product's features, and the reasons for pass use among clients who had used them.Respondents were chosen randomly from the pool of clients every month over 18 months, stratifying each month by credit officer and loan type. 5In total, 575 phone surveys were completed for 457 different clients, representing 18% of all clients in the study sample.Phone surveys were made on average six months after loan disbursement.
Lastly, we conducted an in-person follow-up survey.This survey was brief (median survey duration was 34 minutes) and took place at clients' businesses or homes around ten months (sd=2 months) after the loan disbursement.Since loans were disbursed over time, the survey was conducted on a rolling basis to ensure comparable duration post loan disbursement.Respondents were asked about loan repayment behavior and a set of business and household outcomes.We achieved a response rate of only 69%, we fail to reject a difference in levels across experimental arms (69% response rate for both treatment and control groups) and we fail to reject compositional differences in who is reached (p-value of 0.67 for the pooled treatment and 0.65 and 0.14 for a specification that separately tests compositional selection by the two treatment arms Flex→Flex and Standard→Flex; see Appendix Table 5 for full results). 6Appendix Figure 1 summarizes the timeline of the experiment and related data collection.

Take-up
Figure 1 reports that the 6,685 standard loan offers led to 1,893 disbursed loans (28%) while the 1,925 flexible loan offers led to 582 disbursements (30%).Panel B of Appendix Table 3 shows that the difference in disbursement rates by type of credit offers is not statistically significant (p-value is 0.53).Among applicants, a negative credit assessment was the most common reason for a loan not being disbursed. 5The target sampling rate was initially set to 20% of clients for the first three months of the experiment and later lowered to 5%, subject to a minimum of two calls in each offer-loan type combination in a given month. 6Locating clients in the urban setting of this study was difficult.Clients frequently move the location of the business or place of residence and immediate neighbors are not always willing to provide information about clients' whereabouts.A team of enumerators continually rotated through the different neighborhoods with a list of target respondents and attempted phone contacts to schedule interviews.
Overall, the application outcome and eligibility process was similar for both groups (p-value of joint test is 0.67).
Panel C of Appendix Table 3 shows the take-up rates by recruitment modality.Door-to-door promotions and financial events had similar take-up rates of just over 20% of interested potential clients while over half of potential clients who came to the branch ended up with a loan.In all three recruitment modalities take-up rates were similar for standard and flexible offers.

Selection Effects on Observables
This lack of differences in take-up rates between the offers of standard and flexible loans suggests we are unlikely to see differential composition of clients across the two groups (if one assumes that the addition of flexibility is a free-disposal feature, and hence does not lower take-up rates for any set of individuals).Table 1 compares loan characteristics (from the administrative data) and client and business characteristics (collected by credit officers at the time of the loan application) between borrowers that accepted flexible and standard loan offers.Column 5 reports the p-values of a test of equality of means in columns 1 and 3 and shows that only one difference out of 18 is statistically significant at the 5% level (client's age).It also reports the p-value of an F-test of joint equality for loan characteristics (p-value= 0.81), client characteristics (p-value= 0.37), business characteristics (p-value=0.81) and all characteristics combined (p-value=0.79).We conclude there is no evidence of differential selection on observables using a wide range of observable characteristics.

Selection Effects on Unobservables
We next examine selection on unobservable characteristics by focusing on the use of flexible passes described in Figure 1 and Appendix Table 6.About a third of flexible clients used a pass at any point during the loan (Appendix Table 6, column 1), compared with only 2% among standard loan clients (column 2). 7his take-up rate compares to a 31% flexible contract take-up among clients with standard contracts who were offered a flexible contract in Barboni and Agrawal (2023), with 56% of takers exercising their flexibility option.Our take up rate is lower than the 63% of flexible credit borrowers who used a "voucher" to postpone payments in Battaglia et al. (2023).
While most clients who used a pass at all used only one pass, 40% of such clients used a pass more than once.Flexible loan clients used 0.60 passes on average, roughly evenly split across extension-type passes that added to the maturity of the loan and no-extension type passes where the skipped principal had to be paid within the original loan duration.The limited pass use is consistent with only 8% of flexible credit clients using the maximum allowed number of passes.
Columns 5 and 6 of Appendix Table 6 compare pass use among clients initially offered the standard loan which later switched to a flexible loan to clients offered initially the flexible loan.P-values of a test of equality of means in column 8 show only one difference out of 9 (number of no-extension passes used) is statistically significant at the 5% level.
We conclude that there is no evidence of substantial differential selection on unobservables, at least in pass use.Below we will show similar null results for default rates.We thus pool across initial offers and focus henceforth on the effect of the contract, comparing borrowers of the standard loan and flexible loan (irrespective of the initial offer).This lack of selection refutes the idea that there are many profitable entrepreneurs who reject the standard loan but would accept the flexible loan. 8This result contrasts, however, with the finding of Barboni and Agarwal (2023) that individuals who accept a flexible loan are more financially sophisticated and have considerably more income volatility.Why is there no selection in our case?Data from the lender phone survey of clients indicate that lack of information cannot be an explanation.Panel A of Appendix Table 7 reports almost all flexible credit clients (98%) understood the use of passes.Unlike Barboni and  Agarwal (2023) that required a month-long lag between communication and actual use of the pass, our lender's passes could be used immediately and thus borrowers maybe were more subject to temptation or procrastination in repaying the loan.In addition, unlike most other studies that introduce flexibility, our sample consisted exclusively of new clients who were perhaps less financially disciplined or had on average weaker internalized repayment norms (both because they have not had as much experience learning to repay loans, and because the lending process has not yet filtered out borrowers who are predilected to default).
In addition, the limited use of passes at the start of the loan is not consistent with the idea that flexible credit clients want to use the product to make larger initial investments.Instead, clients might be reacting to business opportunities as they arise or to unexpected negative shocks to business or household finances.
Appendix Figure 2 shows pass use over time.Since not all loans have the same duration, we graph pass use against the proportion of time elapsed in the loan instead of the number of months elapsed.Pass use is lowest on average in the very first months of the loan's duration, increasing until about a quarter of the loan's duration when it reaches its highest point.While anecdotal evidence suggests that some loan officers may have advised clients not to use passes early on, perhaps because of the lower skipped amount 8 One could argue that no selection effects would be detected if the sample excluded applicants interested in the flexible loan but not the standard loan, that is, it only included applicants interested in the standard loan, and if takeup conditional on initial interest was only determined by borrower eligibility (leaving no room for increased demand for the flexible credit).While part of our sample is recruited from branch visits (19% of initial flexible credit applications were made during branch visits) and from financial events (25% of initial applications), where perhaps prospective borrowers approached the lender only knowing about the standard loan, we still find no selection effects when we focus on the sample recruited during door-to-door promotions for which no loan information was provided before revealing the randomized offer type (see Appendix Table 8).In addition, overall take-up conditional on initial interest was only 28% and only about a third of initial applications were rejected because of a negative credit assessment, leaving plenty of room for take-up effects due to the flexibility features.
or due to portfolio risk concerns, pass use still peaks at the first quarter of the loan duration. 9The proportion of extension passes increases over time as clients have less remaining time to repay the skipped balance within the original loan duration.
We report the reasons for pass use given by clients in Panel A of Appendix Table 7.10 Forty-one percent report using the pass to make an investment in the business and separate qualitative data indicates that these business investments include making use of an opportunity for discounted bulk buying of inputs, financing inputs for a large customer order and covering lost revenue from temporarily closing the business for renovations.Dealing with shocks is another important reason why clients use passes ---44% of flexible clients in the phone survey sample who used a pass did so to deal with a personal or family calamity while 19% used a pass to deal with business problems.
Appendix Table 7 Panel B reports client satisfaction using data from the lender phone survey.To keep answers comparable across treatment arms, questions about satisfaction were asked before questions about pass use.While most borrowers feel confident about repaying their loan five months after disbursement (p-value of t-test of equality between flexible and standard loan borrowers is 0.51), borrowers of the flexible loan are 7 percentage points more likely to report higher quality of service from FMSD.Among the reasons given for good service, the flexibility of the product was the only one that was statistically significantly different at conventional levels (p-value 0.00).

Default, Business, and Stress-Related Treatment Effects
We estimate the average treatment effect of a flexible contract relative to a standard contract, pooling across initial offers as discussed above.Since the probability of assignment to a flexible credit offer changed during the experiment (see Figure 1 and Appendix A), we adjust the standard estimation equation to avoid potential bias from correlation of client characteristics with the assignment probability.Following Gibbons, Suarez Serrato, and Urbancic, (2019), we estimate treatment effects separately for the two periods and calculate a weighted average based on the two periods' sample frequencies.Formally, we estimate the following regression equation for client i: where Ti is an indicator for assignment to a flexible contract and R1 is an indicator for receiving an offer in the initial period, Yi is the dependent variable.We include as additional controls the pre-intervention value of the dependent variable when available.β1 and β2 capture the effects of receiving a flexible contract for clients who received offers in the early and late recruitment periods respectively.We then estimate the average treatment effect by averaging the estimates for β1 and β2, proportionally to each period's sample size.
First, we examine repayment behavior, default rates and loan renewal.Table 2 Columns 1 and 2 report outcomes from the administrative data for borrowers of the standard and flexible contract respectively.Panel A reports the raw outcomes while Panel B reports the residuals after regressing default outcomes on the 18 observable characteristics from Table 1 for the standard contract group (with first-stage R 2 values ranging from 0.07 to 0.10).
Regardless of the panel used, the flexible contract group has 3 and 2 percentage points higher proportion of the principal in default 3 and 12 months after maturity, respectively.Column 3 of Table 2 reports the p-value of equality of means and shows this increase in default is statistically significant (p-value<0.01).Despite these default results, we see no statistically significant differences in the share of borrowers who have missed a due payment (i.e., not counting skipped payments from a pass as missed) or the rate of loan renewal.We return below to this pattern of results when exploring the repayment behavior over the course of the loan.
Columns 4 and 5 report the means of the default outcomes in column 2, separating by whether the initial offer was flexible (column 4) or standard (column 5).Column 6 assesses the selection effect by reporting the p-value of the difference in means between columns 4 and 5.As with the comparison using observable characteristics in Table 1 or the use of passes in Appendix Table 5, none of the differences in either Panel A or B is statistically significant.Finally, column 7 reports the difference between borrowers of the standard contract in column 1 and borrowers of the flexible contract in column 5, all initially offered the standard contract.Since we find no selection (column 6), column 7 is similar to column 3 as overall differences in outcomes are attributable solely to differences in the contract.
Next, we examine repayment behavior over the course of the loan to shed additional light on the mechanisms for the results in Table 2.In each graph of Figure 2 we plot the outcome mean among standard borrowers (dashed line), the same mean plus the flexible credit coefficient based on regressions at each point in time (solid line) and the associated pointwise confidence intervals (dotted line).For reference, we also show the rate of pass use over time in a bar chart as in Appendix Figure 2. Again, we use the share of the loan maturity elapsed to account for the variation in loan lengths across the sample and we use the original loan maturity at loan issue to keep flexible and standard contract groups comparable.For additional technical details, see the notes at the bottom of Figure 2.
We document the following repayment patterns.The differential default between flexible and standard loans only appears after the end of the original maturity (Figure 2a).Flexible credit borrowers miss scheduled payments at the same rate as standard credit borrowers during the original loan period (when pass use does not count as a missed payment) but are significantly more likely to miss payments thereafter (Figure 2b).The cumulative rate of ever missing a payment is slightly higher for flexible borrowers, but the difference is not statistically significant at any point during or after the end of the original loan maturity (Figure 2c).Flexible borrowers repay a lower fraction of the principal amount throughout the original maturity, as both extension and no-extension passes are used, and this gap does not close after the end of the original loan period (Figure 2d).
We can draw the following conclusions: First, since default only appears after the end of the original loan period (Fig. 2a and b), only extension passes (rather than no-extension passes) are associated with negative repayment behavior.Second, because the share of flexible borrowers that ever missed a payment is not different compared to that of standard borrowers (Fig. 2c), but the share of flexible borrowers who miss a payment after the end of the original period increases (Fig. 2b), we conclude that the flexible borrowers driving the difference in default rates by missing scheduled payments after the original loan period also missed payments during the original loan period.Third, the lack of treatment effects on loan renewal is consistent with the repayment behavior above as the set of borrowers driving the additional default (only statistically significant at the end of the loan cycle) were already behind on their loans and likely to be ineligible for a follow-on loan.
Next, we examine business, financing and stress-related outcomes using the follow-up survey (Table 3  and 4).Column 1 reports the ATE described in Equation 2. There are no impacts on key outcomes such as sales, profits, or investment (Table 3).Column 6 reports the p-value of a difference in volatility (std.deviation) in sales and profits between the Flexible and Standard Contract groups, but none of the differences is statistically significant.Borrowers of the flexible loan appear to have slightly more businesses and to have started a secondary business.A new enterprise typically is an indication of risktaking, but of course could also be a diversification strategy, and thus we are not able to infer whether the increase in secondary businesses is indicative of flexible lending making risk-taking more palatable for the entrepreneurs.
Table 4 reports no changes in additional business or financing outcomes and no change in an overall loanrelated stress index, although borrowers of the flexible loan report thinking less about loan repayments and a decrease in anxiety in the days prior to loan payment deadlines.Table 4 also reports no change in a general stress index, though flexible loan borrowers report being less nervous or stressed.
In sum, we find no changes in revenues or profits in follow-up data collected about 10 months after loan disbursal but an increase in defaults among the Flexible Contract group.This group also reports lower stress and higher client satisfaction.Using Causal Forests to test for heterogenous treatment effects (Athey, Tibshirani, and Wager 2019; Chernozhukov et al. 2020), we do not find evidence that effects vary systematically as a function of important client or business characteristics pre-loan disbursement, such as gender, sales or household expenses.

Conclusion
We study a flexible lending contract for first-time microcredit borrowers.We find that while flexibility was used by clients, there are no differences in the characteristics or take-up rates between flexible loan borrowers originally offered the flexible loan (Flex→Flex group) and those offered the standard loan (Std→Flex group).This lack of selection effects suggests the lender would not grow its client base much if it offered flexibility to new clients (although longer-term results, particularly given positive customer feedback, may indicate that more time and spreading of information would lead to stronger client acquisition).In addition, first-time borrowers of the flexible loan had higher default rates and limited downstream benefits.These results can help explain why lenders offer rigid loans, particularly to new clients.
Our sample includes only new clients.This is both a feature and a wart.Studying new clients is important for a more complete understanding of credit markets for small-scale entrepreneurs as they may lack experience with managing simultaneous cash flows and repayments.On the other hand, we cannot compare our results to those of more veteran borrowers studied in the literature discussed above, and our study's context differs from that of prior work in more than one way (see Appendix Table 1 for an overview of some salient features).We believe the comparison of new versus veteran clients is an important line of inquiry for future research on loan contract flexibility.
The epilogue to the study is indicative of a broader challenge.The lender viewed the use of passes as a simple way of handling repayment difficulties and introduced a modified version of the flexible loan for non-study loans.Crucially, however, only credit officers (and not clients) decided when to use a pass and clients were not made aware of the feature ahead of time.Pass use thus became merely a tool for credit officers to adjust default and pursue enforcement and refinancing when needed.
While such a policy may have its merits, it deviates from the goal of a product that allows borrowers, fearful of default, to take on higher-risk higher-return investments with the comfort of knowing they have some flexibility to repay.We see these results as motivating, for both lenders and researchers, to continue to learn more about how products can better "match cash flows" both with respect to timing and risk.Notes: The graphs use FMSD's monthly administrative data to show treatment effects over the course of the loan.Graphs show the mean in the standard loan group at a given point in time (dashed yellow lines), mean of the standard loan group plus flexible contract treatment effect (solid blue lines) including 95% confidence intervals (dotted blue lines).Regressions are based on monthly data.Since loans in our sample differ in length, we show the share of loan duration elapsed on the horizontal axis rather than months.We use a loan's original length to make flexible loans and standard loans comparable.We round the share of the loan elapsed to the nearest increment of 0.025 with linear interpolation for values in bins between data points for each loan.We use a similar process for the pass-use bar graph.Pass-use bar graphs are based on bins of 0.0833 (1/12) that match the modal 12-month loan length.

Marketing and
Figure   FMSD also provide us with the loan's repayment status over time.We see exactly when the lender classifies a loan as delinquent and subsequently "cancels" the loan.The initial administrative data did not report payments after the loan was "canceled."The bank offered these delinquent borrowers an opportunity to restructure their remaining debt after their loan got "canceled."These borrowers had the opportunity to continue paying outstanding principal with reduced interest and fees.Since these payments occur after the bank cancels the loan, we do not observe whether delinquent borrowers continue paying their loans.Furthermore, we do not observe the reduction in interest and fees that the bank offers to customers as a part of the re-structuring.In order to properly record payments that delinquent borrowers make, we obtain lender administrative data on every payment that our sample borrowers make from 2015 to 2019.By merging the payment records with the rest of the administrative data, we observe payments that delinquent borrowers make after the bank canceled their loan and reaches out with a restructured proposal.We subtract the payments that borrowers make after their loan gets canceled from their last outstanding principal before the bank "canceled" their loan.The payment data do not distinguish between principal, interest, and fees.We only observe the payment that each borrower makes in a given month.Since our outcome variable is outstanding principal, subtracting a payment that includes interest and fees gives us an incorrect calculation of a borrower's outstanding principal.This means that the 238 borrowers who repay after the borrower cancels their loan have an incorrect measure of outstanding principal for the time after their loan gets canceled. (1) (  Notes: Regressions with sales, expenses, and profit as the outcomes (rows 1-3) control for the baseline value of the outcome.Outcomes are winsorized at the top and bottom 1 percent.Columns 1, 2, and 6 show results for regressions with Flexible Contracts (pooled Std-Flex and Flex-Flex) as the treatment group and Standard Contracts as the control group.Index of Business Activities (row 6) was constructed by calculating a primary component analysis (PCA) score of the outcomes in rows 1-5.The same process was done to constuct the indices in rows 7 and 8, one for activities for the client's primary business and the other for activities for the client's non-primary business(es).P-values of the tests of equality of means in column 2 are based on regressions that control for treatment assignment probability; for additional details, see Section 4. P-values of tests of equality of standard deviations in column 6 were calculated using a randomization inference procedure in which we ran 2,000 independent iterations of randomization into flexible or standard contracts and calculated the difference in standard deviations of an outcome between the flexible and standard contract groups in each iteration.The p-value indicate the proportion of simulations in which the absolute value of the difference in standard deviations was smaller than the difference in standard deviations in our actual experimental assignment.

Interest Rates
At the start of the study in October 2015, FMSD charged between 36% and 42% interest rate with a 70-30 split, respectively.Over time, the share of loans with 42% increased so that by the end of the study in March 2017 all loans were charged 42% interest rate.

Randomization
During the first five months of the intake process (corresponding 15% of offers) the randomization procedure assigned one third of potential clients to a flexible credit offer and the remaining two thirds to a standard credit offer.From month six onward the proportion assigned to receive a flexible offer was reduced to 20% to increase the sample allocated to the standard-standard treatment group (i.e., those who both were offered and received the standard loan).The initial treatment assignment probability was set to balance the selection and impact hypotheses, but after initial analysis and feedback from the bank and observing the process, we decided to increase power for the impact research question relative to the selection question.
For the first-stage randomization, in the beginning of the experiment, until May 2016, we carried out the randomization by using a combination of potential clients' initials, day of offer and time of offer.Quasirandom, traceable characteristics of the interaction with the prospective client were used to prevent the possibility of promoters or credit officers gaming the system and adjusting offers based on client characteristics.We subsequently changed the randomization procedure to both make compliance monitoring easier logistically, given the large number of offers that were being made, and to allow for stratification of offers.The revised first-stage randomization procedure worked as follows: We assigned a fixed set of offers to each staff member that participated in promoting loans, either promoters, credit officers or front office staff, with the number of assigned offers depending on their role in the process (e.g. more offers to promoters, who had more promotion contacts).The offer sets were divided into blocks of offers.For each staff member, the size of the blocks was calibrated to approximately match the expected number of offers made during a two-week period.Randomization was then stratified by staff-member and block.The offer sequences were pre-loaded into the phones used for prospective client registration and the order of offers as registered was periodically checked by project staff against the pre-defined order of offers.

Section B. Marketing Script and Promotional Brochure
Good morning Sir/Madam.I am visiting you from Fundacion Mario Santo Domingo.
Today we are offering loans to people who wish to strengthen or expand their business.
Any type or size of business can access our offer.
Note for the enumerator: Before continuing make sure the person passes the following filter questions.
• OWNS THE BUSINESS

• BUSINESS HAS BEEN FUNCTIONING FOR 6 MONTHS • DOES NOT HAVE A BAD REPORT IN DATACREDITO
• IS NOT OVERINDEBTED • ALSO: make sure the client does not have an active loan application.

Did the person pass the filter?
No → The person does not qualify for our loans.Move on to the next client.
Are you interested in hearing about the offer that we have available today?
No → The person is not interested.Move on to the next house.

If the offer is for a NON-FLEXIBLE loan:
If the offer is for a FLEXIBLE loan: ORANGE KIVA: Kiva NON-FLEXIBLE loan offer Type of interest: 3% monthly.(36% annually.)WITHOUT the right to postpone installments RED KIVA: Kiva FLEXIBLE loan offer Type of interest: 3% monthly.(36% annually.)WITH the right to postpone installments Is the interviewee interested in the offered product?
Not interested → Thank you very much for your time.We are leaving all the information in this flyer.If you have any questions you can call us on the phone numbers listed there.Have a good day.
Wants to proceed with the application → Thank you very much for your interest.To continue with the loan process I need you to give me some personal information.With these, the loan officer can get in touch with you over the course of the week, and if everything goes well, in 2 or 3 days you will have your loan.
Will think about it → I will leave this flyer with all the information.If you do decide to access our loan, you can call the loan officer whose number is on the flyer.However, to access the offer we gave you today I would need to take some personal information.Cada oportunidad de aplazar su cuota de capital se conocerá como pase.Aplazar el pago del capital de su cuota mensual ayuda al crecimiento de su negocio y mejora su capacidad de pago.Este producto está diseñado para fortalecer su negocio y así aumentar sus beneficios.
2. Llame al asesor de la FMSD con anticipación al pago de su cuota del mes y explíquele las razones por las que va a utilizar el pase.Él le indicará el monto a pagar.
4. Aproveche el valor del capital de la cuota para responder a la situación por la cual solicitó el pase.
 NO impedirá que reciba un crédito de mayor valor en el futuro.

Test of selection on observables
Notes: 1.The study includes collateral-free loans provided to women with monthly group meetings (Dabi), and larger collateral-backed debt loans to both female and male borrowers without group meetings (Progoti).2. The loan period was set to 3 years for credit line clients and 1, 1.5 or 2 years for term loan clients.3. Line of credit size decided by loan officers depending on characteristics of the borrower and their business.Papers featured: BA: Barbosi and agarwal (2022); BGM: Battaglia, M., S. Gulesci, and A. Madestam (2021); BGK (in bold italics ): Brune, L, X. Giné and D. Karlan (this paper); FPPR: Field, E., R. Pande, J. Papp, and N. Rigol (2013); SK: Shonchoy, A. and T. Kurosaki (2014) "Impact of Seasonality-adjusted Flexible Microcredit on Repayment and Food Consumption: Experimental Evidence from Rural Bangladesh" IDE Discussion Paper No. 460.AAK: Aragon, F. M., A. Karaivanov, and K.  Krishnaswamy (2020)."Credit lines in microfinance: Short-term evidence from a randomized controlled trial in India."Journal of Development Economics, 102497."Liability" refers to the liability structure.IL refers to individual liability where the borrower is resposible for the repayment of the loan.JL refers to joint liability "Lag to use it?"refers to whether the use of the pass had to be communicated to the lender with a lag of an instalment period or more."Selection into flex contract?"refers to whether a choice between the Flexible and Standard Contract was given to the borrower.FMSD also provide us with the loan's repayment status over time.We see exactly when the lender classifies a loan as delinquent and subsequently "cancels" the loan.The initial administrative data did not report payments after the loan was "canceled."The bank offered these delinquent borrowers an opportunity to restructure their remaining debt after their loan got "canceled."These borrowers had the opportunity to continue paying outstanding principal with reduced interest and fees.Since these payments occur after the bank cancels the loan, we do not observe whether delinquent borrowers continue paying their loans.Furthermore, we do not observe the reduction in interest and fees that the bank offers to customers as a part of the re-structuring.In order to properly record payments that delinquent borrowers make, we obtain lender administrative data on every payment that our sample borrowers make from 2015 to 2019.By merging the payment records with the rest of the administrative data, we observe payments that delinquent borrowers make after the bank canceled their loan and reaches out with a restructured proposal.We subtract the payments that borrowers make after their loan gets canceled from their last outstanding principal before the bank "canceled" their loan.The payment data do not distinguish between principal, interest, and fees.We only observe the payment that each borrower makes in a given month.Since our outcome variable is outstanding principal, subtracting a payment that includes interest and fees gives us an incorrect calculation of a borrower's outstanding principal.This means that the 238 borrowers who repay after the borrower cancels their loan have an incorrect measure of outstanding principal for the time after their loan gets canceled.

Appendix
(1) ( loan from institution other than FMSD Number of business improvement activites (out of 12) Hours worked per day Panel B: Loan-related stress outcomes Panel C: General stress outcomes Thinks about loan repayments at least once per week Anxiety rises in the days prior to loan payment deadlines Had problems with loan payments in last year Not confident that loan will be repaid Cra 45 # 34-01 Piso 2 Tel.3710707 Ext.48046 Cartagena: El Bosque, Calle 21 # 47-95 Tel.6930010 Ext.48209 Bogotá: Av.Calle 26 # 68C-61 Oficina 612 Tel.6070707 Ext.48305 Para que lleve control de su crédito flexible: 2: Contract Effects on Default Outcomes Over Time Notes: P-values based on regressions that control for treatment assignment probability; for additional details, see Section 4.

Table 2 : Contract and Selection Effects in Default Notes
: In Panel B, we obtain residuals after regressing default outcomes on the 18 observable characteristics from Table1for the Standard Contract group, controlling for treatment assignment probability.P-values based on regressions that control for treatment assignment probability; for additional details, see Section 4. Note on outstanding principal: 238 borrowers have a slightly incorrect version of the outcome variable "remaining outstanding principal."Outstanding principal comes from FMSD administrative data.

Table 3 : Effects on Main Business Outcomes (Survey Evidence 10 Months Disbursement) Comparing Flexible Contract (Flex→Flex & Std→Flex) to Standard Contract (Std→Std)
: Columns 1 and 2 show results for regressions with Flexible Contracts (pooled Std-Flex and Flex-Flex) as the treatment group and Standard Contracts as the control group.Outcomes in rows[3] and [4]are winsorized at the top and bottom 1 percent.P-values based on regressions that control for treatment assignment probability; for additional details, see Section 4. Notes

¡Es muy fácil aprovechar los beneficios de su crédito flexible! ¡No dude en aprovechar las ventajas de su crédito flexible!
Note: There are two pass uses that are not shown which occurred beyond the end of the graph's range for the horizontal axis (1 in bin 1.333-1.417and 1 in bin 1.750-1.833).

Table 8 : Contract and Selection Effects in Default for Borrowers Who Got Loans From Door-to-Door Salespeople Notes
: In Panel B, we obtain residuals after regressing default outcomes on the 18 observable characteristics from Table1for the Standard Contract group, controlling for treatment assignment probability.P-values based on regressions that control for treatment assignment probability; for additional details, see Section 4. Note on outstanding principal: 238 borrowers have a slightly incorrect version of the outcome variable "remaining outstanding principal."Outstanding principal comes from FMSD administrative data.