Policy Research Working Paper 11097 The Role of Market Frictions in Demand for Prepaid Electricity Megan Lang Development Economics A verified reproducibility package for this paper is Development Research Group available at http://reproducibility.worldbank.org, April 2025 click here for direct access. Policy Research Working Paper 11097 Abstract Prepaid electricity contracts lower enforcement costs but borrowed and demand for the credit was inelastic; however, may burden consumers, particularly when market frictions the line of credit did not change average demand for elec- are present. This paper presents the results of a randomized tricity. Detailed administrative data reveal that consumers control trial where 2,000 randomly selected, rural Rwan- primarily used the line of credit to lower transaction costs, dese consumers were offered a line of credit for electricity suggesting that rural consumers highly value convenience. payments. The line of credit lowered liquidity constraints The results highlight potential Pareto improvements from and transaction costs. Twenty percent of consumers more flexible prepaid contracts. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at mlang@worldbank.org. A verified reproducibility package for this paper is available at http://reproducibility. worldbank.org, click here for direct access. RESEA CY LI R CH PO TRANSPARENT ANALYSIS S W R R E O KI P NG PA The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Role of Market Frictions in Demand for Prepaid Electricity Megan Lang∗ Keywords: Prepaid electricity, liquidity constraints, transaction costs, solar. JEL Codes: D12, O12, O13 ∗ Economist, World Bank Development Research Group. Correspondence may be sent to mlang@worldbank.org. This project was supported with a grant from the International Growth Centre, project number 38432. The opinions and conclusions expressed herein are solely those of the authors and should not be construed as representing the opinions or policies of the sponsoring agencies or the World Bank. The author gratefully thanks the IGC Rwanda team, particularly Derek Apell, as well as Ethan Ligon, Jeremy Magruder, Elisabeth Sadoulet, Alain de Janvry, Marco Gonzalez-Navarro, Aprajit Mahajan, Edward Miguel, Julia Seither, Carly Tractman, Gabe Englander, Bob Rijkers, participants at the UC Berke- ley development lunch series and development workshop series, participants at the AERE annual meeting, seminar participants, and anonymous reviewers for their suggestions. Thanks to Andrew Kent, Anthony Mabonga, and Philip Sendi for their collaboration and assistance with implementation. This study is pre- registered with the AEA registry AEARCTR-0004887). It has approval from the Rwandan National Ethics Committee (No. 0018/RNEC/2019) and IRB approval in the U.S. (U.C. Berkeley CPHS 2018-02-10770). 1 Introduction Efforts to achieve universal energy access face significant fiscal challenges. Utility providers often operate at heavy fiscal losses, in part due to high costs of enforcing customer contracts and collecting payments (Trimble et al., 2016). Prepaid electricity contracts dramatically lower enforcement costs (Jack and Smith, 2020). However, prepaid contracts impose mul- tiple burdens on consumers. Consumers must align cash flows with demand for electricity, be attentive to electricity consumption to avoid running out at critical times, and incur transaction costs each time they prepay. In a context where prepaid access is the norm, I examine how offering a more flexible contract affects consumer welfare and firm profitability. The contract simultaneously lowers liquidity constraints and transaction costs. Transaction costs may suppress demand for electricity by making it inconvenient to purchase. Liquidity constraints may make it difficult for consumers to align cash flows with demand for electricity. If consumers face transaction costs, liquidity constraints also increase costs by limiting the amount consumers can buy in each transaction. I study liquidity constraints and transaction costs using a randomized control trial (RCT) with 11,570 pay as you go (PAYGo) solar customers in rural Rwanda. PAYGo solar is a criti- cal tool in the push for universal access to electricity, with over 11 million households gaining access through PAYGo solar products since 2020 (Reynolds and Paixao, 2023). PAYGo con- sumers make a small down payment to have a solar home system installed, then “pay as they go” using mobile money to use the system. Unlike on-grid contracts, PAYGo customers purchase access time, or time they can use the system, rather than kilowatt hours. Access time runs down continuously: consumers cannot save it for later. Consumers get remotely locked out of their system when they run out of access time, preventing them from using electricity until they buy more. Consumers face time-based transaction costs for each pur- chase because they travel to mobile money agents to make purchases. The average consumer in my sample travels thirty-four minutes, one way, to reach the nearest mobile money agent 1 and makes 3.3 transactions per month. Continuous rundown with transaction costs creates trade-offs even when consumers are not liquidity constrained. Buying in small quantities incurs repeated transaction costs, but buying in bulk may lead to wasted access time. Liquidity constrained consumers must align cash flows with demand. As such, PAYGo contracts may create burdens for many types of consumers when combined with liquidity constraints and/or transaction costs. I randomly offer a short-term line of credit for PAYGo access time to 2,000 current customers over a four-month period. The line of credit may relax short-term liquidity con- straints or provide credit at a lower price. It lowers transaction costs because consumers call the solar company to borrow rather than traveling to a mobile money agent. Combining experimental variation with detailed administrative data allows me to identify the effects of offering the line of credit along multiple dimensions of consumer behavior:overall demand for solar, default on the PAYGo contract, the number of transactions consumers make, the size of each transaction, and the number of watt hours they consume. Such detailed data speak to the mechanisms underlying changes in consumer behavior. Administrative data also reduce measurement error relative to survey data because the firm relies on the data for their day-to-day operations. It improves external validity because the experiment occurs under real-world conditions. The line of credit is welfare-enhancing for consumers. Twenty percent of consumers borrow, with the number of borrowers growing linearly over the course of the experiment. Demand for the line of credit is highly inelastic. However, it leads to no statistically or economically significant change in average demand for electricity among any subset of con- sumers. Transaction sizes, watt hours consumed, and default rates also remain unchanged. However, I observe a significant reduction in the number of in-person transactions consumers make, suggesting that consumers use the line of credit to lower transaction costs. My results indicate that the line of credit is Pareto improving: companies can offer more flexible prepaid contracts that improve consumer welfare without reducing revenues or increasing default. 2 My experiment quantifies the importance of common market frictions in rural electricity markets. Evidence to date shows mixed and often limited benefits from electrification for rural consumers (e.g., Khandker et al. (2009), Bensch et al. (2011), Dinkelman (2011), Lip- scomb et al. (2013), Khandker et al. (2014), Chaplin et al. (2017), Lenz et al. (2017), van de Walle et al. (2015), Burlig and Preonas (ming)). My results suggest that liquidity constraints are not significantly limiting intensive-margin demand for electricity among consumers who take-up PAYGo solar. Although consumers value tools to lower transaction costs, doing so does not change demand for electricity. It follows that these market frictions do not explain low demand for electricity among the rural consumers in my sample. The experiment also provides some of the first evidence on the role of market frictions for intensive margin demand for electricity, rather than the extensive margin adoption decision. Recent work has highlighted the importance of information frictions in the adoption of solar technology in low-income settings (Mahadevan et al. (2023), Alem and Dugouan (2022)), as well as the role of credit and affordability (Grimm et al. (2020), Lee et al. (2020), Wong et al. (2022)). Understanding the mechanisms shaping electricity use decisions after adoption is critical for designing fiscally sustainable electrification policies. My work further speaks to the welfare implications of PAYGo contracts, extending the literature beyond the prepaid on-grid contracts (Jack and Smith (2015),Jack and Smith (2020)). My work is most closely related to Bonan et al. (2023), who find that making salient the flexibility in traditional PAYGo contracts without providing planning assistance increases rates of default, particularly for consumers with high cognitive constraints and those facing the highest transaction costs. By contrast, my experiment finds positive effects of flexibility. Combined, these results suggest that providing flexibility that addresses relevant market frictions may relieve some of the cognitive burdens that prepaid electricity contracts impose on consumers. Beyond their relevance for debates on electrification, consumer responses to the line of credit underline the burden imposed by time-based transaction costs. Economic theory as- 3 serts that the opportunity cost of time is low for low-income consumers (Becker, 1965). However, the RCT demonstrates that consumers are willing to pay for the line of credit to save time. The lower bound on the value of time implied by consumer behavior in the RCT is high relative to consumption spending and other estimates in similar contexts (e.g., Cook et al. (2016), Wondemu (2016), Kremer et al. (2011), Jeuland et al. (2010), Whittington et al. (1990)). This suggests that the transaction costs associated with electricity payments peri- odically impose substantial burdens on consumers. It highlights complementarities between energy access and financial inclusion, as robust mobile money ecosystems lower transaction costs for a wide range of goods and services. 2 Background: Pay as You Go Solar in Rwanda Pay as you go solar is an important steppingstone in the global push for universal elec- trification. In areas where expanding the grid is costly or households cannot afford grid connections, solar home systems provide reliable access to basic electricity. In Rwanda, the national electrification plan calls for 48% of rural households to be electrified using off-grid solutions in the short- to medium-term given the high costs of extending the grid and limited anticipated demand for electricity (ess, 2018). PAYGo is a way for households to finance access to electricity. Consumers choose to adopt a solar home system that is bundled with high-efficiency appliances such as light bulbs, radios, portable torches, phone chargers, or televisions.1 Once a consumer has selected appliances, they make a down payment and have the solar panels, a battery for storing electricity, and all appliances installed.2 After the solar home system (SHS) has been installed, consumers “pay as they go” to pay off the SHS. The solar company sets a daily rate for solar access time based on the appliances 1 Figure A1 summarizes appliance ownership. All systems include at least two lights with phone charging ports. 2 The down payment amounts to 3-5% of the total value of the PAYGo contract in my sample, but contracting terms vary between solar firms. Terms are fixed: consumers cannot make a larger down payment to reduce their daily rate. 4 included. Consumers prepay for access time using mobile money. Once a consumer has purchased access time, they have access to their SHS for the duration of the purchased period.3 When access time runs out, the solar company remotely locks the consumer out of their SHS, preventing them from using it until they buy more access time.45 If the consumer does not purchase access for over thirty consecutive days, the solar company may repossess the SHS.6 Remote lockout and a credible threat of repossession make PAYGo solar contracts highly enforceable. In my context, paying for access time 85% of the time results in the consumer owning their system after approximately three years.7 Although PAYGo solar contracts carry low upfront costs, ongoing use costs are substan- tial. Median weekly consumption spending on food and other essentials is RWF 6680 (USD 7.44) for a rural household of four (National Institute of Statistics of Rwanda (2017))8 . The most basic systems used by 75% of consumers in my sample cost between RWF 910 and 1330 (USD 0.98–1.47) for a week of access. This implies that the solar home system would comprise around 13%–20% of consumption spending for the median household.9 The flexibility of PAYGo contracts is limited by continuous rundown and transaction costs. Access time runs down continuously regardless of how much a consumer uses their solar home system. Consumers cannot choose to delay the start of their purchased time, and they cannot choose to voluntarily shut off their system to save access time for later. Traveling to a mobile money agent is a transaction cost. In phone surveys with two separate samples of PAYGo solar customers in Rwanda, I asked how long it takes to reach the nearest mobile money agent. Responses are an upper bound on time costs since other tasks might 3 The battery is large so consumers are rarely constrained by the capacity of the system or variations in cloud cover. 4 The solar panel continues charging the battery while the consumer is locked out, but the consumer cannot access the electricity generated. Any electricity generated after the battery is fully charged is lost. 5 See Gertler et al. (2024) for applications of remote lockout that go beyond electricity contracts. 6 Although it is theoretically possible that a household could pay for just one day per month, transaction data show that this is extremely rare. 7 My sample frame excludes consumers who paid off their system during the experiment. 8 See Figure A2 for the full distribution 9 This is comparable to what households would spend for on-grid electricity, although grid connections have higher upfront costs and are often unavailable in rural areas (Blimpo et al., 2018). 5 bring consumers close to a mobile money agent. In the combined samples, the average time is 50 minutes and the median time is 30 minutes one-way (see Figure A3). In the sample of treated consumers, the mean time is 34 minutes and the median is 30 minutes. Consumers vary widely in the number of transactions they make each month: the median is 3 and the 75th percentile is 5, but the 98th percentile goes up to 15 transactions per month (see Figure A4). Consumers can reduce transaction costs by depositing funds in their mobile money ac- count and later using them to buy access time. In practice, 66% of consumers report visiting a mobile money agent five out of the last five times they paid for solar and 78% report visiting four of the last five times.10 This pattern is likely the result of limited mobile money take-up.11 Transactions conducted with mobile money are free, but consumers have to pay withdrawal fees to convert mobile money into cash. With low mobile money take-up, consumers cannot use mobile money to buy most goods so withdrawal fees render it less liquid than cash. In addition, the mobile network is unreliable at times in rural areas, which makes paying from home less certain than transacting with an agent. As such, time-based transaction costs are prevalent for PAYGo consumers. Liquidity constraints may exacerbate the challenges of PAYGo contracts. Consumers need to ensure that they have cash on hand to buy access time when they value it the most. The median transaction size prior to the experiment is 6.25 days. Such small transactions are consistent with liquidity constraints or waste minimization. Taking all features of the setting together, the perishability of access time combined with transaction costs and liquidity constraints creates stark trade-offs for consumers. They need to align cash flows with their demand for solar while minimizing transaction costs. In the next section, I describe the product I use to ameliorate these trade-offs. 10 Figure A5 shows complete survey results. Figure A6 indicates that trips to the mobile money agent are not driven by lack of knowledge: nearly 80% of consumers report that they know how to use mobile money. 11 A 2018 report by the World Bank found that only 31.1% of adults in Rwanda had mobile money accounts (WBG, 2018). 6 3 Experimental Design I partner with a solar company in Rwanda to offer existing PAYGo customers a line of credit for PAYGo access time. With the line of credit, consumers can call the solar company and request to borrow up to one or two weeks of access time. The line of credit may alleviate liquidity constraints by allowing consumers to purchase access time when they do not have cash on hand. It may reduce transaction costs because consumers call the solar company to borrow rather than traveling to a mobile money agent. For instance, a consumer can call to borrow five days of access time, allow their system to turn off, then go to the mobile money agent to repay the borrowed time and to purchase additional access time. Making the same set of transactions without the line of credit would require two trips to the mobile money agent. Alternatively, a customer may borrow enough access time to maintain access until the next market day, effectively lowering transaction costs by allowing them to visit the mobile money agent when they were already planning to go to town. Consumers can decouple cash flows from their demand for electricity and better time trips to the mobile money agent. The line of credit operates within the PAYGo enforcement environment. If a consumer’s borrowed time runs out before they have repaid, they are remotely locked out of their system. When a consumer makes a payment after borrowing, the funds first go to repaying the borrowed time plus a flat fee. This ensures that consumers cannot default on the line of credit without defaulting on their PAYGo contract. Any funds that are left after repaying the borrowed time go to pre-paying for additional access time. The repayment scheme deliberately mimics short-term loans for airtime on mobile phones, which were common at the time of the experiment. Consumers can borrow as many times as they like over the course of the experiment. The experiment is designed to study consumer responses to the line of credit, holding appliance ownership constant. In the medium- to long-term, household demand for electricity may increase as households adopt more appliances. The experiment cannot capture these changes; however, I observe few system upgrades. There is virtually no correlation between 7 consumer tenure with the solar company and the daily rate, indicating that daily rates tend to remain static over relatively long time horizons (see Figure A7). The solar company agreed to randomize access to credit at the level of the individual consumer, effectively a household, and to cross-randomize the terms of the line of credit along three dimensions: borrowing limits, price, and repayment time frames. Half of treated consumers could borrow up to seven days of access time and half could borrow up to fourteen days, although all consumers could choose a voluntarily lower borrowing limit than the one randomly assigned. Half of consumers faced a 10% flat fee on borrowed days and half a 2% fee.12 Finally, as a risk mitigation measure for the firm, half of consumers were informed that they would lose access to the line of credit if they did not repay within one week of the borrowed access time running out. The other half did not face a time limit, but as usual got remotely locked out of their system when they ran out of access time. All customers in the Northern and Southern provinces of Rwanda who had signed a PAYGo contract with the solar firm at least 90 days prior to the start of the experiment are in my sample. In total, the firm offered the line of credit to 2,000 randomly selected customers.13 For context, I combine responses to a phone survey conducted with 1,229 randomly selected PAYGo solar customers in 2019 with the a nationally representative survey of Rwandan households last conducted in 2016–2017 to construct a wealth index (National Institute of Statistics of Rwanda (2017)). I find that PAYGo customers are wealthier than the average rural Rwandan household.14 I stratify treatment based on demand in the 90 days before the experiment. My measure of demand is the utilization rate (UR): the proportion of days a consumer has purchased system access. I create four strata: 0-30%, 30%-65%, 65%-80%, and 80%-100%. My sample has 1,325 consumers in the first stratum, 1,404 in the second, 986 in the third, and 7,855 in the top stratum. I treat 500 in each stratum.15 12 10% is comparable to rates for borrowing airtime. 13 See Table A1, which shows the treatment and control group are balanced on covariates. 14 Figure A8 shows the distribution of wealth scores. 15 Figure A9 shows the distribution of pre-experimental demand along with lines denoting the strata. I 8 3.1 Timeline and Data The solar company marketed the line of credit starting on October 14, 2019. They called each treated customer to explain the terms of the line of credit and how to access it. All consumers also received an SMS message containing details of the line of credit. After completing the initial round of marketing calls, the solar company attempted to call every treated consumer again to complete a short survey and to further educate customers about the line of credit starting in late November.16 The second round of calls asked consumers who had not borrowed why they had not. Roughly half had forgotten how or had forgotten that they could, so the second round of calls served to improve understanding and awareness. Consumers could use the line of credit through February 14, 2020, at which time all treated consumers received an SMS message informing them that the program had ended. My primary source of data is the administrative records of the solar company. The dataset of loan requests and repayments allows me to observe borrowing and estimate price elasticities for the line of credit. I use administrative data to conduct initial balance checks, estimate changes in the monthly utilization rate, and examine underlying mechanisms. I use data from a phone survey with treated customers conducted by the solar company partway through the experiment to measure the distance from the nearest mobile money agent.17 selected the strata using the firm’s rough line for profitability: an 80% utilization rate. I roughly divided the remaining distribution into groups of equal sizes to ensure coverage across the distribution of pre-experimental demand. 16 Marketing reached 64% of consumers in the 0%-30% stratum on the phone during at least one of the rounds of calls. They reached 86%, 90%, and 96% of consumers in the 30%-65%, 65%-80%, and 80%-100% stratum. Differences across strata occur because many consumers in the lowest strata are close to default and thus have weak incentives to answer calls from the solar company. The sample is not balanced across those who were reached with marketing and those who were not; however, this is simply another stage of take-up. 17 See Appendix B for complete details on the phone survey. 9 4 Results For all outcomes that I observe over time, I estimate reduced-form average treatment effects over the four months of the experiment using the specification yit = βtmtit + γi + γjt + ϵit , (1) where yit is the outcome of interest for consumer i in month t, tmtit is a binary variable indicating treatment status (either pooled across cross-randomized treatments or disaggre- gated by treatment arm), γi is a consumer fixed effect, and γjt is a stratum by month fixed effect.18 I also present pre-registered estimates of heterogeneous average treatment effects by strata.19 4.1 Demand for the Line of Credit I first examine take-up for the line of credit, the simplest revealed preference measure of consumers’ value for lowering transaction costs and liquidity constraints. There is strong demand overall for the line of credit. Figure 1 shows that 20% of treated consumers borrow at least once over the course of the experiment, in line with take-up of un-targeted microcredit (Banerjee et al. (2015)). Take-up is slightly lower among consumers who face a repayment time limit but not significantly so. The average borrower takes 1.8 loans and the average loan size is 7.9 days. Growth in the number of new borrowers is roughly linear throughout the experiment, suggesting that the total number of borrowers would have continued to grow had the experiment continued (see Figure A12). Figure 1 shows that take-up is significantly higher in the top three strata than in the lowest stratum of pre-experimental demand. This pattern is driven by a large number of consumers in the lowest stratum who do not make any transactions during the experiment.20 18 I focus only on the experimental period due to COVID-19 restrictions that went into effect soon after. 19 Figure A10 and Figure A11 show additional pre-registered heterogeneity. 20 Eighty percent of consumers in the control group and 82% of treated consumers in the lowest stratum 10 Excluding these consumers, borrowing rates are 23% in the lowest stratum, in line with estimates in other strata. Descriptive correlates of borrowing are consistent with liquidity constraints and transac- tions costs. Borrowing is negatively correlated with pre-experimental transaction sizes and positively correlated with the number of remote lockouts in a month (see Table A2). It is also positively correlated with the distance to the nearest mobile money agent. Variation in the flat fee charged on borrowed access time allows me to estimate the price elasticity of demand for the line of credit. I estimate log(DaysBorrowedi ) = η log(F eei ) + δ log(F eei ) ∗ T imeLimiti + βj Xij + γs + ϵi . (2) j log(DaysBorrowi ) is the logarithm of the number of days borrowed during the experiment by consumer i. F eei is the fee faced by consumer i. As pre-specified, I only consider η the price elasticity to isolate consumer responses to price; however, results are nearly identical if I omit the interaction between the repayment time limit and the fee.21 I control for daily rates to account for differences in fees arising from different daily rates given that the randomly assigned fees are a percentage of the amount borrowed. The vector of controls also includes pre-experimental demand and the randomized borrowing limit. γs is a strata fixed effect. Figure 2 shows elasticities estimated for all consumers and heterogeneously by strata. Demand is highly inelastic. I can reject elasticities of -0.5 across all strata and elasticities of -0.1 pooling the sample. I cannot reject perfectly inelastic demand for any strata, although consumers with the lowest pre-experimental demand qualitatively have more elastic demand. High take-up of the line of credit coupled with highly inelastic demand suggests that the line of credit is welfare-enhancing for consumers. Correlates of borrowing suggest that both market frictions may play a role. Given this evidence, I turn to the effects of the line of do not purchase access time during the experiment. Consumers with zero transactions are likely waiting to have their systems repossessed. 21 Note that I cannot study changes in the price elasticity given that I only observe two price points. 11 credit on demand for solar access time. 4.2 Demand for Solar Access Time Figure 3 shows average treatment effects on demand for access time pooled, disaggregated by treatment arm, and by strata. I cannot reject that the average treatment effect on demand for solar is zero for any group of consumers. The estimates are precise: I can reject that the line of credit changed demand by more than 1.5 days per month, on average, for consumers in any strata. For consumers in the top two strata, this means that I can rule out effects of more than 2.5% of demand. Results are robust to alternative specifications (see Table A3 and Table A4). Effects on utilization are consistent across all cross-randomized treatments. Although around half of borrowers assigned to the repayment time limit do not repay on time, and subsequently lose access to the line of credit, effects among consumers facing the repay- ment time limit are not significantly different from those among consumers who had no time limit.22 The same is true for the borrowing limit: effects are not significantly higher for consumers with a higher borrowing limit. Although initially surprising, only 45% of borrow- ers assigned to the high borrowing limit ever borrow more than 7 days, consistent with a limited role for liquidity. In line with inelastic demand for the line of credit, fee levels do not significantly change effects on solar demand. Null effects on demand could result from some consumers increasing demand and others defaulting more. This is a particular concern if certain consumers are present biased. Changes in purchase timing when consumers borrow could lead present biased consumers to borrow but then procrastinate on repayment (Laib- son (1997), O’Donoghue and Rabin (1999)). In the most extreme cases, such procrastination could lead to higher rates of PAYGo contract default and repossession. Null effects on demand for access time are not driven by higher default rates on PAYGo contracts. Figure 4 shows effects pooled and by strata. For each group, I also show effects 22 Extensive results by time limit, which were included in the pre-analysis plan, are available upon request. 12 separated by borrowing limit, as higher borrowing limits are more likely to lead to default among present biased consumers. Default rates do not change significantly among any group, although results are imprecise because default is a low-frequency event. Qualitatively, default rates among treated consumers in the lowest strata decline by roughly 50%, on average. Pooling across all strata, the effect on default is -0.3 percentage points, around 8% of the control group mean. Higher borrowing limits do not lead to higher rates of default. This suggests that present bias is not a first order concern for the line of credit, likely because consumers still face remote lockout. The pre-registered analysis presents a puzzle. Many consumers borrow and have inelas- tic demand for the line of credit, indicating that consumers highly value the line of credit. However, the line of credit does not increase demand for solar access time. What explains consumers’ high value for the line of credit? The next section presents the results of unreg- istered analysis examining the mechanisms underlying the main effects. 4.3 Mechanisms I examine the number of in-person transactions, meaning all non-borrowing transactions, average transaction sizes, and the number of watt hours consumed when systems are switched on to understand the mechanisms driving consumer responses to the line of credit.23 If transaction costs drive demand for the line of credit, I should observe a reduction in the number of in-person transactions when consumers can borrow and consolidate trips to the mobile money agent. Observing an increase in transaction sizes is consistent with liquidity constraints combined with transaction costs shaping demand for electricity: consumers prefer to make larger purchases to lower total transaction costs, but liquidity constraints prevent them from doing so. Finally, the number of watt hours used on days when a system is switched on speaks to whether the line of credit allows consumers to better target purchases 23 Since most consumers report traveling to the mobile money agent nearly every time that they buy access time, I interpret the number of non-borrowing transactions as a close proxy for the number of in-person transactions made. 13 and waste less access time. If a consumer purchases access on a day when they have low demand for electricity, they may use fewer watt hours than on high-value days. The top panel of Figure 5 shows that the line of credit significantly reduces the number of in-person transactions. Effects are largest for consumers in the top two strata but are still negative, though not significantly so, for consumers in the second stratum. The results indicate that consumers in the top two strata reduce the number of in-person transactions they make by 0.25 transactions per month, a 6–8% reduction. The bottom two panels of Figure 5 show that there are no significant changes in trans- action sizes or the number of watt hours consumed: I can reject changes of more than 6% in transaction sizes and changes of more than 0.5% in watt hours consumed when pool- ing across all strata and treatment groups. Null effects on watt hours consumed indicate that consumers were already targeting their purchases well. This is also consistent with null effects on transaction sizes. If consumers are carefully targeting their purchases, then transacting in large amounts with a higher risk of waste will not be optimal. Taken together, the results suggest that consumers value the line of credit for the conve- nience it provides. They use it to maintain demand while making fewer in-person transac- tions, lowering transaction costs. In the PAYGo setting with continuous rundown, liquidity constraints do not appear to be significantly limiting electricity demand. The results show that rural consumers are willing to substitute money for time savings. I observe the time it takes each treated consumer to reach the nearest mobile money agent from phone survey data. Since consumers may combine trips to the mobile money agent with other tasks, this is an upper bound on the time-based transaction costs a consumer would incur to pay for solar. I divide the total amount paid in loan fees by the maximum possible time savings: the number of loans multiplied by the time it takes to reach the nearest mobile money agent. This yields a lower bound on the value of time saved using the line of credit. Conditional on borrowing and responding to the phone survey the median value of time is RWF 266 per hour (USD 0.30) and the mean is RWF 760 (USD 0.84) (see Figure A13). 14 Median daily per capita consumption expenditures among rural respondents are RWF 647, with a first quartile of RWF 441 and a third quartile of RWF 825 (National Institute of Statistics of Rwanda, 2017). Even accounting for positive selection into PAYGo contracts, the value of time implied by consumer responses to the line of credit indicates a high willingness to pay for time savings. This suggests that rural consumers periodically face significant transaction costs when paying for access time. Twenty-five percent of loans were requested after sunset and 10% in the hour before sunset, potentially indicating that the line of credit is particularly valuable when consumers decide at the last moment that they want to purchase access time for the evening. Having the flexibility to adjust purchase timing appears to be highly valued by rural consumers. 5 Discussion The results point to three lessons for policy. First, consumers have a high willingness to pay to lower transaction costs, underscoring that even rural, low-income consumers value convenience. This emphasizes the importance of investing in infrastructure and systems like mobile money that lower a range of transaction costs. Second, the results underline the importance of understanding consumer incentives and behavior when designing policies. Although rural consumers often face general liquidity constraints, continuous rundown in PAYGo contracts render liquidity second-order. It follows that increasing demand for access time would require lowering prices or increasing the benefits of electrification, not simply expanding access to liquidity. Finally, the remote lockout technology used in PAYGo con- tracts can be leveraged to write even more flexible contracts that benefit consumers without increasing default costs for firms. Policies that encourage firms to design contracts that bet- ter suit low-income, rural consumers will be central to fiscally sustainable policies for rural electrification and development. My results have limited applicability to other electricity contracts. The line of credit allows consumers to use access time before they pay for it, similar to postpaid contracts. 15 However, the line of credit can still leverage the low-cost enforcement build into prepaid contracts. PAYGo solar contracts are also structured differently than on-grid prepaid contracts. On-grid contracts charge per kilowatt hour while PAYGo solar contracts charge for access time, allowing consumers to adjust on margins that are unavailable under PAYGo contracts. For instance, liquidity constrained consumers with prepaid on-grid contracts may limit appli- ance use, whereas PAYGo consumers make a binary decision about access time. Relatedly, PAYGo access time is strictly perishable. This lowers the importance of liquidity constraints but increases transaction costs relative to on-grid prepaid contracts. Even so, inattention to electricity use or misaligned incentives for energy use between household members as docu- mented in Jack et al. (2018) could render prepaid on grid electricity “perishable” in practice, creating similar trade-offs. Finally, PAYGo solar contracts bundle appliances with electric- ity provision, creating an unusual relationship between appliance adoption and demand for electricity. In an on-grid contract, adopting additional appliances does not change the cost of using existing appliances. PAYGo contracts build the cost of the appliance into the price for access time, increasing the price for existing energy services. This may slow appliance adoption relative to on-grid prepaid electricity contracts, limiting the impact of liquidity constraints in PAYGo contracts. The other relevant dimension of external validity is consumers who have not opted into PAYGo contracts. Since my sample only contains existing customers, I cannot speak to the effect that it may have on adoption. This leaves open the possibility that the most acutely liquidity constrained consumers are not opting into PAYGo contracts and would benefit from the line of credit. 6 Conclusion The rapid growth of PAYGo solar among rural, African consumers makes it a critical prepaid contract to study as governments push towards universal access to electricity. I randomly 16 offer a line of credit to PAYGo solar customers in Rwanda to study the role of two important market frictions: liquidity constraints and transaction costs. The line of credit does not change demand for electricity on average, but consumers value the line of credit because it allows them to reduce transaction costs. The study has two key limitations. First, as previously discussed, the experiment cannot identify the role of liquidity constraints and transaction costs on the extensive margin. Sec- ond, the RCT only speaks to the burdens consumers face when PAYGo contracts interact with liquidity constraints and transaction costs. In practice, many prepaid contracts also impose cognitive costs on consumers that are difficult to observe. Both limitations points to the need for further work on consumer decision-making around electricity adoption and use in rural, low-income settings. These findings strengthen arguments in favor of prepaid electricity contracts but suggest that using more flexible contracts can increase consumer welfare without reducing utility revenues or increasing the likelihood of default. Although the line of credit in my experiment is specific to PAYGo solar, my results highlight the varied ways flexible contracts can improve welfare for rural consumers. 17 References (2018). Energy sector strategic plan. (2018). The little data book on financial inclusion 2018. Technical report, International Bank for Reconstruction and Development. Alem, J. and Dugouan, E. (2022). Learning from unincentivized and incentivized commu- nication: A randomized controlled trial in india. Economic Development and Cultural Change, 71. Banerjee, A., Karlan, D., and Zinman, J. (2015). Six randomized evaluations of microcredit: Introduction and further steps. American Economic Journal: Applied Economics, 7:1–21. Becker, G. S. (1965). A theory of the allocation of time. The Economic Journal, 75:493–517. Benjamini, Y., Krieger, A. M., and Yekutieli, D. (2006). Adaptive linear step-up procedures that control the false discovery rate. Biometrika, 93(3):491–507. Bensch, G., Kluve, J., and Peters, J. (2011). Impacts of rural electrification in Rwanda. Journal of Development Effectiveness, 3(4):567–588. Blimpo, M., McRae, S., and Steinbuks, J. (2018). Why are connection charges so high? an analysis of the electricity sector in sub-saharan africa. No. 8407. Bonan, J., d’Adda, G., Mahmud, M., and Said, F. (2023). Nudging payment behavior: Evidence from a field experiment on pay-as-you-go off-grid electricity. The World Bank Economic Review. Burlig, F. and Preonas, L. (Forthcoming). Out of the darkness and into the light? develop- ment effects of rural electrification. Journal of Political Economy. Chaplin, D., Mamun, A., Protik, A., Schurrer, J., Vohra, D., Bos, K., Burak, H., Meyer, L., Dumitrescu, A., Ksoll, C., and Cook, T. (2017). Grid electricity expansion in tanzania by 18 mcc: Findings from a rigorous impact evaluation. Technical report, Mathematica Policy Research. Cook, J., Kimuyu, P., Blum, A. G., and Gatua, J. (2016). A simple stated preference tool for estimating the value of travel time in rural africa. Journal of Benefit Cost Analysis, 7:221–247. Dinkelman, T. (2011). The Effects of Rural Electrification on Employment: New Evidence from South Africa. American Economic Review, 101(7):3078–3108. Gertler, P., Green, B., and Wolfram, C. (2024). Digital collateral. Quarterly Journal of Economics, 139(3):1713––1766. Grimm, M., Lenz, L., Peters, J., and Sievert, M. (2020). Demand for Off-Grid Solar Electric- ity: Experimental Evidence from Rwanda. Journal of the Association of Environmental and Resource Economists, 7(3):417–454. Jack, K., Jayachandran, S., and Rao, S. (2018). Environmental externalities and free-riding in the household. Working Paper 24192. Jack, K. and Smith, G. (2015). Pay as you go: Pre-paid metering and electricity expenditures in south africa. American Economic Review Papers and Proceedings, 105(5):237–241. Jack, K. and Smith, G. (2020). Charging Ahead: Prepaid Metering, Electricity Use, and Utility Revenue. American Economic Journal: Applied Economics, 12(2):134–168. Jeuland, M., Lucas, M., Clemens, J., and Whittington, D. (2010). Estimating the private benefits of vaccination against cholera in beira, mozambique: A travel cost approach. Journal of Development Economics, 91:310–322. Khandker, S., Barnes, D., and Samad, H. (2009). Welfare impacts of rural electrification: A case study from bangladesh. WPS 4859. 19 Khandker, S. R., Samad, H. A., Ali, R., and Barnes, D. F. (2014). Who Benefits Most from Rural Electrification? Evidence in India. The Energy Journal, 35(2). Kremer, M., Leino, J., Miguel, E., and Zwane, A. (2011). Spring cleaning: Rural water impacts, valuation, and property rights institutions. Quarterly Journal of Economics, 126:145–205. Laibson, D. (1997). Golden eggs and hyperbolic discounting. Quarterly Journal of Eco- nomics, 112:443–478. Lee, K., Miguel, E., and Wolfram, C. (2020). Experimental evidence on the economics of rural electrification. Journal of Political Economy, 128(4):1523–1565. Lenz, L., Munyehirwe, A., Peters, J., and Sievert, M. (2017). Does Large-Scale Infrastructure Investment Alleviate Poverty? Impacts of Rwanda’s Electricity Access Roll-Out Program. World Development, 89:88–110. Lipscomb, M., Mobarak, A. M., and Barham, T. (2013). Development Effects of Electrifica- tion: Evidence from the Topographic Placement of Hydropower Plants in Brazil. American Economic Journal: Applied Economics, 5(2):200–231. Mahadevan, M., Meeks, R., and Yamano, T. (2023). Reducing information barriers to solar adoption: Experimental evidence from india. Energy Economics, 120. National Institute of Statistics of Rwanda (December 2017). Integrated household living conditions survey(2016-2017), version 0.1 of the public use dataset. O’Donoghue, T. and Rabin, M. (1999). Doing it now or later. American Economic Review, 89:103–124. Reynolds, O. and Paixao, S. (2023). Insights from the january-june 2023 sales and impact data. Technical report, GOGLA. 20 Trimble, C., Kojima, M., Perez Arroyo, I., and Mohammadzadeh, F. (2016). Financial viability of electricity sectors in sub-saharan africa: Quasi-fiscal deficits and hidden costs. Working Paper 7788. van de Walle, D., Ravallion, M., Mendiratta, V., and Koolwal, G. (2015). Long-Term Gains from Electrification in Rural India. The World Bank Economic Review, page lhv057. Whittington, D., Mu, X., and Roche, R. (1990). Calculating the value of time spent collecting water: Some estimates from ukunda, kenya. World Development, 18:269–280. Wondemu, K. (2016). The economic value of time: Evidence from africa. South African Journal of Economics, 84:230–244. Wong, J. C. Y., Blankenship, B., Harish, S., and Urpelainen, J. (2022). Increasing microso- lar technology adoption: Efficacy of vouchers, cash transfers, and microfinance schemes. Energy Economics, 110. 21 7 Figures and Tables Figure 1: Use of the Line of Credit by Cross-Randomized Treatment and Stratum Note: Proportion of treated consumers who use the line of credit at least once over the course of the experiment, by cross-randomized treatment and stratum. Dark bars show 95% confidence intervals calculated using naive standard errors. Light bars show 95% confidence intervals that correct for multiple hypothesis testing by controlling the false discovery rate using the adaptive step-down procedure from Benjamini et al. (2006). Black point estimates indicate pre-registered analyses. Grey point estimates indicate analyses that were not pre-registered, but which I include in the multiple hypothesis adjustments for this family of outcomes to be conservative. 22 Figure 2: Price Elasticity of Demand for the Line of Credit by Stratum Note: Price elasticity of demand for the line of credit estimated using Equation 2, by stratum. Bars show 95% confidence intervals. 23 Figure 3: Average Treatment Effects on Demand for Solar by Cross-Randomized Treatment and Stratum Note: Average treatment effects on the utilization rate: the propostion of time a customer’s solar home system is switched on. Bars show 95% confidence intervals calculated from standard errors clustered at the level of the individual consumer. Triangles show mean utilization rates in the control group during the experimental period. Figure 4: Average Treatment Effects on Default by Stratum and Borrowing Limit Note: Average treatment effects on the PAYGo default rate, by stratum and borrowing limit.Bars show 95% confidence intervals calculated from robust standard errors. Grey points show estimates that were not pre-registered. Triangles show average default rates in the control group during the experimental period. 24 Figure 5: Average Treatment Effects on Potential Mechanisms by Stratum Note: Average treatment effects on the number of in-person transactions made each month (top panel), the average transaction size (middle panel), and the average number of watt hours consumed on days when a system is swtiched on (bottom panel). Bold, wide bars show 95% confidence intervals calculated from standard errors clustered at the level of the individual consumer. Light, narrow bars show 95% confidence intervals implied by applying the adaptive step-down procedure from Benjamini et al. (2006) to the full set of estimate effects on the three potential mechanisms represented here. Triangles show mean values among the control group during the experimental period for each mechanism. Note that none of the estimates presented in this figure were pre-registered. 25 A Appendix - for Online Publication A.1 Additional figures Figure A1: Appliance ownership Note: Proportion of consumers in my sample who own each of the most common appliances included with the solar home system. Note that all systems come with lights. Therefore, “Only lights” is the proportion of consumers with systems that only provide lights. All other appliance categories are not mutually exclusive. 26 Figure A2: Weekly per capita consumption expenditures among non-urban households in Rwanda Note: Histogram of weekly expenditures among non-urban households, as reported in the 2016–2017 Integrated Household Living Survey. Vertical lines denote the 25th, 50th, and 75th percentiles. Figure A3: Consumer Travel Times to Nearest Mobile Money Agent, One-way Note: Travel times are self-reported during one of two phone surveys. Respondents report the time it takes to reach the nearest mobile money agent using whatever method of transport they would typically use. 27 Figure A4: Pre-experimental purchasing behavior by stratum Note: The top figure shows box plots of mean transaction size (in days of access time) in the 3 months before the experiment by stratum. The bottom figure shows box plots of the number of transactions in a month in the three months before the experiment by stratum. The bottom of the box is the 25th percentile and the top is the 75th percentile, with the bold horizontal line showing the median. Vertical lines show the maximum value 1.5 times above the 75th percentile and below the 25th percentile. Dots show points lying outside this range. 28 Figure A5: Self-Reported Use of Mobile Money Agents to Buy Solar Note: Responses to the question, “of the last five times you purchased access time, how many times did you need to visit the mobile money agent to do so?” Respondents are randomly selected customers of the solar company. Figure A6: Self-Reported Knowledge of Using Mobile Money to Buy Solar Note: The figure shows the proportion of consumers in each stratification bin who answer ”yes” to the question, “If you had mobile money already in your account and you wanted to use it to pay for solar, do you know how you would do that?” 29 Figure A8: Distribution of Wealth Among Rural Households in Rwanda Note: The figure shows the distribution of wealth indices for a nationally representative sample of rural households in Rwanda using data from the 2016–2017 Integrated Household Living Conditions Survey (National Institute of Statistics of Rwanda (2017)). The red line is mean wealth for the nationally representative sample. The blue line is mean wealth for consumers in my experimental sample. I use the following variables to construct the wealth index: ubudehe category (a government-assigned category designed to summarize the socio- economic status of a household), roof material, wall material, floor material, primary source of electricity (if any), primary source of light, whether or not the household is connected to the national grid, and weekly energy expenditures. Figure A7: Customer tenure and daily rate Note: Average tenure (in days) of consumers with each daily rate paid for system access time. Flat trends illustrate that few consumers upgrade their systems over time to include more appliances. 30 Figure A9: Distribution of Demand Prior to the Experiment Note: The 90-day pre-experimental utilization rate is the proportion of time a consumer purchased access to solar in the 90 days immediately prior to the experiment. Vertical black lines show where I bin consumers for the purpose of stratification when I assign treatment. Figure A10: Elasticity by Pre-Experimental Variance in Use Note: Price elasticity of demand for the line of credit estimated using Equation 2, by pre-experimental variance in watt hours used. Bars show 95% confidence intervals. 31 Figure A11: ATEs on Solar Demand by Pre-Experimental Variance in Use Note: Average treatment effects on the utilization rate by pre-experimental variance in use. Bars show 95% confidence intervals calculated from standard errors clustered at the level of the individual consumer. Triangles show mean utilization rates in the control group during the experimental period. Figure A12: Growth in Take-up of the Line of Credit Over Time Note: The blue line shows total number of loans taken by all treated consumers in each week of the experiment. The red line shows the total number of first time borrowers each week of the experiment. 32 Figure A13: Distribution: Lower bound on the value of time implied by borrowing Note: I estimate the value of time for each borrower by dividing the total amount they pay in fees over the course of the experiment by the number of loans taken multiplied by the hours they report it takes to reach the nearest mobile money agent. Self-reported time to reach the nearest mobile money agent is an upper bound on the time-based transaction costs associated with paying for solar, as consumers may be able to combine trips. Using an upper bound for time savings yields a lower bound on the value of time. Figure A14: Dynamic effects: solar demand Note: Dynamic treatment effects on demand for solar electricity. Bars denote 95% confidence intervals calculated using standard errors clustered at the level of the individual consumer. 33 Figure A15: Dynamic effects: in-person transactions Note: Dynamic treatment effects on the number of in-person transactions. Bars denote 95% confidence intervals calculated using standard errors clustered at the level of the individual consumer. 34 Table A1: Balance Table Dependent variable: 90-day Pre UR Daily Rate Mean wH SD wH Tenure Pmt Size (1) (2) (3) (4) (5) (6) Treated −0.00003 2.062 0.179 0.569 −1.438 0.550 (0.002) (2.299) (0.545) (0.355) (5.417) (0.415) Stratum 1 −0.902∗∗∗ 30.657∗∗∗ −35.479∗∗∗ −11.680∗∗∗ 41.608∗∗∗ −0.379 (0.002) (2.634) (0.753) (0.402) (6.135) (0.683) Stratum 2 −0.466∗∗∗ 18.295∗∗∗ −24.125∗∗∗ 1.238∗∗∗ 28.999∗∗∗ −3.406∗∗∗ (0.002) (2.549) (0.579) (0.391) (5.974) (0.424) Stratum 3 −0.230∗∗∗ 13.776∗∗∗ −14.947∗∗∗ 1.554∗∗∗ 14.911∗∗ −3.977∗∗∗ (0.002) (3.016) (0.684) (0.467) (7.136) (0.507) Constant 0.957∗∗∗ 200.231∗∗∗ 52.136∗∗∗ 16.384∗∗∗ 452.124∗∗∗ 13.301∗∗∗ 35 (0.001) (0.962) (0.214) (0.148) (2.267) (0.160) Observations 11,570 11,169 10,702 11,570 11,570 10,596 R2 0.956 0.017 0.273 0.080 0.006 0.011 Adjusted R2 0.956 0.017 0.273 0.080 0.005 0.010 Notes: I report robust standard errors in parentheses. Column (1) shows balance on the utilization rate in the 90 days immediately prior to the start of the experiment. Column (2) shows balance on the price consumers pay for a single day of solar access. Columns (3) and (4) show balance on the average watt hours and standard deviation of watt hours used, respectively, when systems are switched on in the 90 days prior to the experiment. Column (5) shows balance on customer tenure with the firm. Column (6) shows balance on transaction sizes in the 90 days prior to the experiment. 36 Table A2: Borrower Characteristics Dependent variable: Ever Borrow (1) (2) (3) 7 days, 2% 0.032 0.026 0.028 (0.035) (0.005) (0.006) 7 days, 10% 0.044 -0.004 -0.005 (0.035) (0.001) (0.002) 14 days, 2% 0.052 0 0 (0.035) (0) (0) 14 days, 10% 0.044 0 0 (0.035) (0) (0) 7 days, 2%, Time Limit 0.032 0.119 -0.059 (0.035) (0.035) (0.068) 7 days, 10%, Time Limit 0.024 0.034 0.03 (0.034) (0.04) (0.055) 14 days, 2%, Time Limit 0.024 0.033 0.018 (0.034) (0.04) (0.054) Pre-Experimental Demand 0.119 -0.059 (0.035) (0.068) Daily Rate 2e-04 2e-04 (1e-04) (2e-04) Tenure 1e-04 0 (1e-04) (1e-04) Mean Transaction Size -0.004 -0.0051 (0.001) (0.0017) Mean Switch Offs 0.026 0.028 (0.005) (0.006) Mean Use (wH) 0.002 (0.001) SD(Use) (wH) -0.003 (0.002) Time to MM Agent 0.07 (0.027) Constant 0.168 0.032 0.196 (0.024) (0.053) (0.082) N 2000 1651 1149 Notes: Pre-experimental demand is the proportion of time each month a consumer has access to solar. Daily rate is the price the consumer pays for a day of solar in RWF, where higher prices correspond to larger systems that support more appliances. Tenure is the number of months since adoption. Mean transaction size is the average transaction size in days (e.g., normalized by the daily rate). Mean switch offs is the average number of times a consumer is remotely locked out of their system in a month. Mean use and SD(Use) are the mean and standard deviation of watt hours used on days when the system is switched on. Time to MM agent is the self-reported time it takes a consumer to reach the nearest mobile money agent. I report robust standard errors 37 in parentheses. Table A3: Robustness to Alternative Specifications Utilization Rate No. In-Person Transactions (1) (2) (3) (4) (5) (6) DiD Cross Sec- ANCOVA DiD Cross Sec- ANCOVA tion tion Treated -0.0054 -0.0063 -0.0061 -0.1751 -0.3036 -0.3035 (0.0055) (0.006) (0.0057) (0.0447) (0.0574) (0.0569) Control Mean 1 1 1 3 3 3 N 79020 44940 44940 80990 46280 46280 Notes: Average treatment effects on demand for solar and the number of in-person transactions, pooling across all strata and all cross-randomized treatments. DiD denotes the difference in difference specification, which includes individual fixed effects and strata by month fixed effects. Cross-sectional denotes a regression comparing the treatment to the control group using only treatment months, controlling only for strata fixed effects. ANCOVA denotes the cross sectional regression with an additional control for pre-experimental demand. Parentheses show standard errors clustered at the individual level. Table A4: Robustness to Alternative Measure of Demand (1) (2) (3) DiD Cross Sec- ANCOVA tion Treated -0.006 0.0017 0.0018 (0.0048) (0.0039) (0.0039) N 79020 44940 44940 Notes: Average treatment effects on the likelihood of having continuous access to electricity in a month, pooling across all strata and all cross-randomized treatments. DiD denotes the difference in difference specification, which includes individual fixed effects and strata by month fixed effects. Cross-sectional denotes a regression comparing the treatment to the control group using only treatment months, controlling only for strata fixed effects. ANCOVA denotes the cross sectional regression with an additional control for pre-experimental demand. Parentheses show standard errors clustered at the individual level. 38 B Phone survey details The solar company re-called all treated customers roughly a month into the experiment to conduct a short phone survey. The survey contained the following questions. 1. (For those who had not borrowed) We have noticed that you are eligible to request days through [program name], but you have not done so yet. We are curious to know why you have not used [program name] yet - can you tell me about that? (Select multiple) • Didn’t need to have lights on • Had enough money to pay for solar this whole time • Didn’t know how to request days • Forgot they could request days • Didn’t believe [program name] was real • Thinks the fee is too expensive • Was worried they couldn’t repay on time • Doesn’t believe in borrowing to pay for solar/doesn’t like debt • Other 2. Would you like me to tell you about [program name] again? 3. (For those who had borrowed) We have noticed that you have used [program name]! We would like to know more about your experience. When/why did you decide to use [program name]? • Didn’t have money to pay for solar • Didn’t have time to go to a mobile money agent • Prefers to pay after using solar • Decided at the last minute that they wanted solar/didn’t know that they would want the lights on • More convenient • Wanted to test it out/see if it was real • Other 4. About how long does it take you to reach the nearest mobile money agent when you go to pay for solar? 5. How does knowing that you can request days through Tira BBOXX change how you pay for solar? • Doesn’t change anything • Don’t need to buy days as often 39 • Don’t need to buy as many days at once • Don’t always have to go to mobile money agent to buy days • Less stress about negative consequences for missing days • Feel more supported by the solar company • Feel less afraid of the solar company • Other 6. Do you have any questions about [program name]? 7. Initially, we offered you to borrow up to [insert borrowing limit]. At the time, you chose to limit yourself to only being able to request up to [insert chosen borrowing limit]. Can you tell me why you chose a lower limit? • Didn’t understand what it meant to choose a lower limit • Not comfortable borrowing more than that • Worried they would borrow more than they could repay on time • Chose the number of days they typically pay for at once • Other 8. Would you like to change your choice? If yes What would you like your limit to be? 40