Abstract
Combining administrative data on credit, internet penetration and a land reform in Rwanda, this paper shows that the complementarity between technology and law can overcome financial frictions. Leveraging quasi-experimental variation in 3G availability from lightning strikes and incidental coverage, we show that mobile connectivity steers borrowers from microfinance to commercial banks and improves loan terms. These effects are partly due to the role of 3G internet in facilitating the acquisition of land titles from the reform, used as a collateral for bank loans and mortgages. We quantify that the collateral's availability mediates 35% of the overall effect of mobile internet on credit and 80% for collateralized loans.
1 Introduction
The internet has brought lasting changes to contemporary societies, exerting influence on political engagement (Gavazza et al., 2019; Enikolopov et al., 2020; Manacorda and Tesei, 2020; Guriev et al., 2021), employment opportunities (Hjort and Poulsen, 2019; Bircan and De Haas, 2020; Bahia et al., 2023), social media (Braghieri et al., 2022; Fujiwara et al., 2023), and more, as reviewed by Hjort and Tian (2024). At the same time, digitalization has the potential to transform financial markets and institutions by reducing the cost of securing property rights and streamlining the electronic registration and transfer of assets, aligning with seminal works by De Soto (2000), Djankov et al. (2007) and Besley et al. (2012).
The complementarity between technology and law can amplify the impact of property rights reforms and alleviate the classical financial frictions preventing access to finance. Internet has the potential to decrease the cost of participation in these reforms, since the acquisition of information and property-related documentation becomes less expensive. At the same time, technology can foster broader synergies with financial institutions, as they can verify property rights in real time and establish a collateral system, thus enabling individuals to leverage their assets to access credit from banks. Such a transformative process is especially significant in low- and middle-income countries characterized by weak institutional frameworks (La Porta et al., 1998; Besley and Persson, 2011) and severe financial frictions (de Mel et al., 2009; Kaboski and Townsend, 2012; Kaboski, 2024).
This paper offers an empirical contribution to examining this complementarity. We investigate the circumstances under which the combination of technology and property rights expands the range of financial options available to individuals and enhances credit allocation. Initially, we analyze whether the introduction of mobile internet affects the borrowers’ choice of financial institutions between microfinance institutions (MFIs) and commercial banks. Subsequently, we delve into the role of collateral in this reallocation process. Our study is centered on Rwanda, providing an ideal setting due to its extensive administrative data on financial decisions, a reform that introduced a digitally-empowered system of property rights and collateral, along with detailed data on 3G coverage, the primary source of internet connectivity at the time of the reform.
Our empirical investigation requires identifying sources of variation in mobile internet independent of local confounding factors. We leverage two distinctive geographical features that we use to build instrumental variables. First, the heterogeneous distribution of lightning strikes in Rwanda’s municipalities that affects the development of mobile Internet infrastructures, as outlined in Manacorda and Tesei (2020) Second, Rwanda’s topography, which significantly impacts its telecommunications, as demonstrated by Yanagizawa-Drott (2014), also including the highly uneven availability of 3G technology across the country. In this regard, we use the quantification of “incidental coverage” by Björkegren (2019), which measures mobile signal availability based on idiosyncratic topographic features, such as a municipality’s proximity to the electric grid and its location on different sides of a hill’s slopes.1 Importantly, we show that these two sources of internet availability are exogenous to and uncorrelated with income and other predetermined characteristics.
To conduct our empirical tests, we employ extensive administrative data sources from Rwanda. Our primary dataset, obtained from the Credit Reference Bureau, spans from 2008 to 2015 and includes detailed loan-level information from all regulated credit institutions operating within Rwanda. In particular, the registry reports information on all loans granted by both commercial banks and MFIs, which is essential for our research and uncommon in most credit registries. This comprehensive dataset includes details on the loan amount, lending rate, collateral status, and borrower attributes like location, age, gender, and marital status for each loan. In total, our dataset comprises information on 113,897 distinct individuals across 337 municipalities. Each borrower is uniquely identified by a numerical code, facilitating the tracking of their lending engagements over different time periods and across lenders.
We augment this dataset by incorporating data on 3G mobile internet coverage from the GSM Association, which reports annual 3G coverage across all Rwandan municipalities from 2008 to 2015. Notably, the introduction of 3G mobile internet at the country level occurred in 2011, providing a key source of time-series variation in internet connectivity. We also add information on the instrumental variables, which provide a source of cross-sectional variation by municipality. Data on lightning strike frequency is obtained from the Global Hydrology Resource Center, whereas data on incidental coverage are from Björkegren (2019).
Finally, our analysis incorporates information on a digitally-empowered reform that introduced property rights and a collateral system. The Land Tenure Regularization (LTR) program marked the initiation of a land tenure system, accomplished through the “development and maintenance of the digital registry” and the creation of “digital records for over 10 million land parcels” connected to banks through the mortgage registry (World Bank, 2018, 2020). To analyze the implementation of this reform and its relation to the internet, we collaborate with the Rwanda Land Management and Use Authority, building and examining novel data on the distribution of land certificates, the number of transactions, and the share of mortgaged lands at the municipal level.
Our primary finding is that the arrival of mobile internet in Rwanda affected individuals’ financial decision-making and induced a reallocation away from microfinance institutions towards banking, without overall effects on inclusion, echoing the findings of Breza and Kinnan (2021) in India. Our analysis reveals that increased mobile internet availability leads individuals to shift from MFIs to commercial banks, as shown by a 3.4% increase in the probability of having a loan from a commercial bank and a 3.6% decrease in the probability of having an MFI loan in response to a one standard deviation increase in 3G coverage. Additionally, a one standard deviation increase in 3G coverage leads to a 1.1% higher probability that an MFI customer switches to a commercial bank. This transition does not alter overall financial inclusion, defined as the likelihood of individuals receiving a loan from any financial institution.
To understand the heterogeneity behind these effects, we observe that socioeconomic differences between municipalities are central, while individual-specific traits play no discernible role. This underscores the inclusive nature of internet technologies in this context, rejecting evidence of discrimination based on observable characteristics like gender, marital status, and age. Furthermore, our study presents evidence highlighting the positive impact of mobile internet on borrowing opportunities and loan characteristics for individuals accessing bank loans. Borrowers with access to mobile internet tend to have larger outstanding loan amounts without adverse effects on interest rates. For switchers from MFIs to commercial banks, especially those in the upper tail of the interest rate distribution, we observe a significant reduction in the lending rate.
After verifying that mobile internet induces an expansion in banking at the detriment of microfinance, we investigate the mechanism at the heart of this change: mobile internet promotes the acquisition of land titles, which provides access to collateral and the possibility to borrow from banks. Our first piece of evidence is that the implementation of the land reform is larger, faster, and more effective in municipalities with a higher mobile coverage. We find that a one standard deviation increase in 3G coverage leads to a 16 percentage points higher likelihood that the municipality experiences a disbursement of land titles exceeding 50% of the total in the first year of the reform, and an increase by 14% of the total number of transactions.
Using additional data from the land registry, we verify that lands are not only claimed more promptly in area with internet access, but are also used to gain access to the financial system. The share of mortgaged lands increases significantly with the presence of mobile internet: a one standard deviation higher 3G coverage leads to a 1.7 percentage point higher share of mortgaged lands. We conclude turning back to the credit bureau data and exploiting the specific information on whether a loan presents a collateral and whether it is a mortgage. Our findings indicate that the probability of receiving a collateralized bank loan or a loan classified as mortgage increases by 0.6 and 0.8 percentage point, respectively, in response to a one standard deviation increase in the 3G coverage.
These results, together with the previous ones, establish that mobile internet contributes both to the expansion of banking and to empowering the availability of land titles as a form of collateral. In the last tests, we quantify the role of collateral as a mediator of the effect of mobile internet on banking. Following the work of Doerr et al. (2022), we employ a Sobel-Goodman (SG) test and assess the mediation effect of land titles on the impact of 3G coverage on the probability of receiving a bank loan. The findings reveal that approximately 35% of the effect of mobile internet on the probability of receiving a bank loan is indirect and facilitated by our proxy of land certificates’ availability, whereas the remaining 65% is directly attributed to mobile internet.2 To provide additional evidence in line with our hypothesis, we repeat our estimation using as dependent variable the probability of receiving a collateralized loan and find that the direct effect of mobile internet decreases to 20% and is borderline significant, while the effect mediated by the land titles accounts for 80% of the total impact. This quantification points to the pivotal role of collateral in securing a mortgage and steering individuals from MFIs toward commercial banks.
Our findings are robust to a wide range of tests, changes in the sample, and model specification. We replicate our analysis at a more aggregated bank-municipality-year level and find that the results remain unaffected. Then, we test different definitions of the dependent variables, the main predictors, and the instrumental variables. Results are valid when we replace the main instrumental variables (lightning strikes and incidental coverage) with pre-determined physical characteristics that are inputs into these variables: the altitude of a municipality and the coverage of wood, water and sand. This delivers similar estimates, with a higher precision and a first stage F-statistic exceeding 100. At the same time, reduced-form estimates for both IV strategies show results in line with our main specifications. The findings are also robust to the inclusion of control variables, aiming to control for municipality time trends, and to the confounding effects of reforms other than the expansion of the 3G, such as the introduction of the Umurenge Savings and Credit Cooperatives (U-SACCOs) and the rolling out of the fiber, which happened in the period covered in our analysis. Finally, our results on the mediation role of land collateral are robust to a falsification test focusing on the availability of the fiber-optic technology and the implementation of a sequential g-estimation procedure that produces comparable quantifications (Acharya et al., 2016).
This paper contributes to the literature on property rights and legal systems by showing how technology can interact with law to facilitate the acquisition and utilization of property rights for accessing financial markets. A seminal work by (La Porta et al., 1998) studies property rights’ legal rules and enforcement, and its impact on stock market participation, in line with Djankov et al. (2007), who emphasizes property rights and the collateral’s pivotal role in credit market development. In this setting, Besley et al. (2012) show that moral hazard matters for the welfare effects of improved property rights and collateral systems. Two studies on Brazil by Ponticelli and Alencar (2016) and Fazio and Silva (2021) offer interesting results showing that improved enforcement affects firms’ access to finance, investment, and size, and that stronger creditor protection affects credit allocation across space. Our research is especially close to a recent paper by Djankov et al. (2020), who analyse comprehensive data on titling and urban development for multiple countries. They observe that novel information technologies drive an improvement in property right protection through a reduction in time and cost. Our paper resonates with this study and identifies the role of technology in promoting both forms of rights through a major decline in their cost of acquisition and enforcement.
The paper also connects with the literature on land titles (Besley and Ghatak, 2010; Besley et al., 2012; Manysheva, 2021). Two recent papers also offers interesting insights related to our work. Montero (2022) studies a land reform in El Salvador showing that cooperative property rights affect specialization, productivity, and worker equity. Acampora et al. (2022) provide a random subsidy to incentive owners to rent out land, which generates long-term effects since the land rental market remains active upon the subsidy expiration. Our work resonates with the findings that lands may be inefficiently allocated and reforms can spur investment and productivity. A strand of this literature directly studies the effects on credit availability. Ali et al. (2017) look at the land tenure regularization program in Rwanda and suggest that it could expand access to finance. Deininger and Goyal (2012) consider a land registry computerization in an Indian state to argue that this reform increased credit supply by facilitating third party access to registry information and reducing banks’ cost of extending credit. We contribute by studying with granular administrative data how the complementarity between internet technologies and a land reform alleviate financial frictions and allow the transition from microfinance to commercial banks. In this respect, this paper aligns with the broader literature on the relevance of collateral for accessing bank credit, as highlighted by recent work of Ioannidou et al. (2022), and the importance of technology in mortgage lending Fuster et al. (2019) and the expanding role of household finance in emerging markets, as reviewed by Badarinza et al. (2019).
Finally, our research contributes to the literature emphasizing the role of new internet technologies in information access and dissemination. Internet technologies, notably mobile internet, impact various societal aspects, from political perceptions (Guriev et al., 2021) to social organizations, protests (Manacorda and Tesei, 2020), and financial outcomes (Gao and Huang, 2020; Hvide et al., 2022). We highlight the indirect impact of mobile internet on information spread in linked markets, aligning with cross-market spillover effects discussed by Bos et al. (2018). In this respect, our paper adds to a growing literature documenting the impact of new IT technologies on banking (Core and De Marco, 2021; Kwan et al., 2024; Mazet-Sonilhac, 2021; Lin et al., 2021; Jiang et al., 2022; Pierri and Timmer, 2022; D’Andrea et al., 2023; D’Andrea and Limodio, 2024).
The rest of the paper is organized as follows. Section 2 presents the institutional setting in Rwanda. Section 3 describes our data and the main empirical specifications. Section 4 presents the results, sheds light on the mechanism behind our findings, and provides robustness checks. Finally, Section 5 concludes.
2 Institutional Setting
2.1 Credit Market in Rwanda
Historically, Rwanda’s credit market faced numerous challenges, including limited access to financing, lack of financial infrastructure, and inadequate regulatory frameworks. The formal banking sector, consisting of commercial banks and other financial institutions, was primarily concentrated in urban areas, leaving a significant portion of the population (about three quarters of the population is rural based), particularly those engaged in agricultural and informal sectors, with limited access to formal credit channels.
To address the financial needs of underserved populations, microfinance institutions emerged as key players in the credit market. MFIs provided small loans, savings accounts, and other financial services to individuals, primarily in rural and peri-urban areas. These institutions played a role in promoting financial inclusion, especially for micro and small-scale entrepreneurs who lacked access to traditional banking services. By leveraging group lending methodologies and maintaining close relationships with borrowers, MFIs helped fill the gap left by formal banks and stimulated economic activity at the grassroots level.
In recent years, Rwanda has made significant strides in strengthening its credit market, fostering financial sector development, and enhancing access to credit for both individuals and businesses. Financial inclusion increased from 48% to 93% between 2008 and 2020, and the share of adult population using services and products from commercial banks increased from 14 to 36% over the same period (Access to Finance Rwanda, 2020). These changes have been driven by a combination of factors, including economic development, policy reforms, and technological advancements.
Formal banks have sensibly expanded their reach. With improved regulatory frameworks and technological advancements, formal banks have been able to extend their services to previously underserved areas. Nowadays, they offer a wide range of financial products and services, including loans, savings accounts, and digital banking solutions, and play a central role in facilitating investments, supporting businesses, and promoting economic growth.
Parallel to the expansion of formal banking, there has been a shift away from microfinance institutions as the primary source of credit for individuals and small businesses. While microfinance played an important role in providing financial services to the unbanked population, the growth of formal banking has gradually reduced the reliance on microcredit. This shift can be attributed to several factors, such as increased awareness and confidence in formal banking systems, improved regulatory frameworks, and the availability of alternative credit options.
In this paper, we show that technological advancements have contributed to facilitating the expansion of formal banking and the transformation of the credit market in Rwanda. In particular, the widespread adoption of mobile internet has made banking products more accessible, convenient, and cost-effective.
2.2 Mobile Internet
In the last two decades, Rwanda has made significant strides in improving its information and communication technology (ICT) connectivity. Following the devastating civil conflict from 1990 to 1994, the country recognized the potential of ICTs as catalysts for economic development and growth. In line with this vision, the “Vision 2020” program was launched in 2000, aiming to transform Rwanda into a middle-income knowledge economy by 2020. ICT was identified as one of the key drivers for achieving this vision (GoR, 2012). This belief in the potential of new technologies has driven strong activism and political commitment in Rwanda’s journey towards ICT connectivity.
The government has taken various steps to establish a foundation for high-speed connections. Initially, the focus was on attracting private-sector investments through the introduction of enabling laws and the establishment of a regulatory authority for the ICT sector. Subsequently, the country privatized its incumbent telecommunications operator and fostered wireless competition, thereby encouraging additional private-sector investments. Concurrently, the government ensured that new technologies were accessible nationwide, not limited to commercially lucrative areas. Finally, the transparent and dedicated public administration played a crucial role in promoting ICT adoption and building trust between businesses and the government (UN-OHRLLS, 2017).
Mobile technology has become pervasive in Rwanda. In 2015, the country ranked third in Africa in terms of the percentage of the population covered by 3G, following South Africa and Lesotho (GSMA, 2015). The Rwanda Utilities Regulatory Authority (RURA) reported 3.8 million mobile internet subscriptions among a population of almost 11.5 million, surpassing the average for the continent and least developed countries. The impressive diffusion of mobile technology can be attributed to several factors, with affordability being a key driver. Rwanda boasts the cheapest price per gigabyte of data in East Africa when considered as a percentage of per capita income. Consequently, it is often regarded as a positive example for low- and middle-income countries transitioning to a modern economy. The country serves as a benchmark for broadband installation, penetration, and successful implementation of ICT investments.
By leveraging these advancements in mobile technology and ICT infrastructure, Rwanda has laid the groundwork for exploring the impact of mobile internet on access to credit. The widespread availability and affordability of mobile internet have opened new avenues for financial inclusion and the development of the banking sector. In the following sections, we delve into the specific effects of mobile internet on banking, highlighting its role in promoting traditional banking and facilitating the transition from microfinance.
2.3 The Land Tenure Regularization Program
In the aftermath of the civil conflict, the large-scale migration of returnees had a dramatic impact on land ownership, escalating tensions and land conflicts, making land reform a crucial condition to ensure social stability and improve land use management and investments. With this background, in 2009 the government of Rwanda launched the Land Tenure Regularization program, a reform aimed to regulate the ownership and control of the lands, enhance land utilization, and promote its efficient management (Abbott and Mugisha, 2015).
In order to successfully complete such a large-scale initiative, the National Land Authority of Rwanda experienced a significant expansion of its activity, collecting and centralizing a massive amount of data on parcels. After allowing for a period of corrections and addressing objections, with a large-scale mobilization of citizens involving approximately 110,000 people, the data was digitized to facilitate the issuance of land titles. By the year 2012, the database had accumulated information on an impressive 10.4 million land parcels.
The LTR program introduced a land tenure system based on the digital registry (World Bank, 2020). This registry played a pivotal role in favoring access to formal credit since land titles could be used as collateral for loans and thus facilitated access to credit for the landowner (Besley and Ghatak, 2010; Acampora et al., 2022; Manysheva, 2021). In line with this argument, in 2013 all financial institutions in Rwanda were connected to the digital registry, which made it easier the tracking and verification of land ownership.
The execution of the land reform highly benefited from the expansion of mobile internet, with digital advertising playing a prominent role. Recognizing the ubiquity of mobile phones as one of the most diffused modes of communication in the country, the National Land Authority initiated an extensive online advertising campaign, further strengthened by the introduction of its Twitter account in March 2011 (Schreiber, 2017). Mobile internet was used in two ways to promote the program and provide timely updates on its progress: directly, being a network where people could search for information and gain knowledge about the land certificates; and indirectly, as a channel through which people could learn about other sources of information such as radio broadcasts, TV programs and local public meetings.
In this paper, we leverage both the role played by mobile internet in promoting land reform and the availability of the digital registry, and show that part of the effect of mobile internet on banking is channeled through a digitally-empowered land collateral. The latter reduces information frictions between borrowers and commercial banks and allows individuals to move away from microfinance institutions towards formal banking.
3 Data and Empirical Strategy
3.1 Data
Our study uses comprehensive loan-level data obtained from the Credit Reference Bureau (CRB), a private credit bureau regulated by the National Bank of Rwanda (NBR). The dataset encompasses loans provided by all supervised credit institutions in Rwanda, including commercial banks, microfinance institutions, and savings and credit co-operatives (SACCOs), which are government-backed financial institutions aimed at enhancing financial inclusion for Rwandan citizens. The loan-level information is reported on a monthly basis, from January 2008 to December 2015, with no minimum loan size requirement. The credit registry data, that we aggregate at the yearly level, offer a highly representative view of the total banking sector loans (Agarwal et al., 2023).
In our baseline analysis, we focus on loans granted to individuals. For each loan, we have access to information such as the loan amount and price, borrower’s location (down to the municipality level), and borrower characteristics like age, gender, and marital status. After cleaning the data, we compile a comprehensive dataset that captures the local currency lending activities of banks and MFIs for 113,897 unique individuals across 337 municipalities.3 Borrowers are identified with a unique numerical code that allows us to track their lending activity over time and across lenders.
To examine the impact of mobile internet on commercial banking, we incorporate data on mobile phone coverage from the GSM Association (GSMA), a global trade association representing the interests of the mobile phone industry. The data, obtained through a partnership with Collins Bartholomew4, provide yearly geolocated information on mobile coverage in Rwanda from 2008 to 2015.5 The dataset distinguishes between the availability of 2G, 3G, and 4G technologies, with 3G coverage accounting for most of the variation over the sample period. Our data allow us to measure the penetration of 3G at a very disaggregated geographical level, ranging between 1 km2 on the ground (for high-quality submissions based on a geographic information system (GIS) vector format) to 15–23 km2 (for submissions based on the location of antennas and their corresponding radius of coverage) (GSMA, 2015). For the purpose of our analysis, we map the adoption of mobile internet at the municipality-year level. Figure 1 presents a visual representation of the coverage of 3G in Rwanda.
To address endogeneity concerns and establish a causal relationship between the 3G technology and access to credit, we construct two instrumental variables (IVs) using external datasets. The first IV is based on the frequency of lightning strikes, which we obtain from the Global Hydrology Resource Center (GHRC) in Huntsville, Alabama. Data on lightning strikes are collected through the spaceborne Optical Transient Detector (OTD) on Orbview-1, and the Lighting Imaging Sensor (LIS) onboard the Tropical Rainfall Measuring Mission (TRMM) satellite, respectively. The LIS detection efficiency ranges from 88% during the night, to 69% at noon. The OTD has an efficiency of 52% during the night and 37% at noon. All the data are top-coded and released in the LIS/OTD Gridded Lightning Climatology Data Collection. For each grid cell, time and lightning are summed and scaled for the sensor precision’s rate providing as outcome daily, monthly, seasonal, or yearly data. We follow the methodology in Manacorda and Tesei (2020) and improve in precision, using a grid resolution of 0.1 instead of 0.5. The left panel of Figure 2 illustrates the map of average lightning frequency by municipality, this time-invariant measure represents the first IV for mobile internet coverage.
The second IV is incidental coverage, and captures the variation in mobile internet based on the combination of geographical attributes with the preexisting electric grid. Rwanda’s topography, characterized by hilly terrain, influences the cost of providing mobile internet to nearby villages. To account for this, we compute a measure of fictitious coverage that would have resulted if mobile operators had built towers along the entire network of power lines. This variable is constructed following the work by Björkegren (2019), and allows us to capture idiosyncratic variation based on the interaction between geography features and the electric grid.6 The right panel of Figure 2 plots the map of average incidental coverage by municipality, which represents the second IV for mobile internet coverage.
To understand the underlying mechanisms behind our findings, we incorporate additional data from the Rwanda Land Management and Use Authority’s land dashboard, which provides information on land certificates, transactions, and mortgaged lands, that we use to study the spillover effects of mobile internet. While data on land titles are available at the district level, we obtain proprietary data at the municipality level from the authority to further analyze the relationship between land collateral and bank credit access.
Lastly, we gather municipality-level data from the National Institute of Statistics (NISR) and the Integrated Household Living Conditions Survey (EICV). These data capture various indicators related to local economic and financial activities, focusing on the period prior to the introduction of the 3G mobile technology (before 2011). By incorporating these municipality-level controls, we aim to reinforce the robustness and comprehensiveness of our empirical analysis.
Table 1 presents summary statistics of the main variables used in our empirical analysis. Panel A examines access to credit, specifically the probability of having an outstanding loan, and separately having a loan with a commercial bank or a microfinance institution, and identifies the switching borrowers. Panel B provides yearly average characteristics of the bank credit relationship: outstanding amount, principal amount, interest rate, and the probability that the loan is secured. Finally, panels C and D present summary statistics for the main predictor, 3G mobile coverage, and its instrumental variables, including lightning frequency and incidental coverage.
3.2 Identification
A concern with the estimates of a model in which we regress credit features on mobile internet is that new technologies are unlikely to be randomly allocated across areas, potentially generating a bias in the estimates of model parameters. In order to deal with this concern, we use an IV strategy that exploits exogenous differential rates of 3G adoption across Rwandan municipalities.
The first instrument is lightning frequency, which leverages the negative correlation between frequent lightning strikes and mobile connectivity. During storms, frequent electrostatic discharges can disrupt the signal transmitted by ground-based antennas, thereby reducing both the supply and demand for mobile phone services. This, in turn, hampers the profitability of technology investments and discourages the adoption of mobile services (ITU, 2003). Therefore, areas with a higher incidence of lightning strikes are expected to exhibit slower adoption of mobile technologies compared to areas with lower lightning frequency (Manacorda and Tesei, 2020). Figure 3, left panel, provides initial evidence supporting this hypothesis in the context of Rwanda. We observe that municipalities below and above the median of lightning strikes had zero 3G coverage before the introduction of the technology in 2011, after which their trends diverge in a monotonic manner.
The second instrument is incidental coverage, and builds on the heterogeneity in the cost of providing mobile internet to different areas based on their proximity to the existing electric grid and the topography of their land. Operating mobile towers connected to the electric grid is more cost-effective than establishing standalone towers. Hence, a tower’s transmitter has a higher probability of being constructed close to the existing grid. However, only places that are visible from a height of 35 meters above the tower are likely to receive mobile coverage, and that depends on the topography of the area. Exploiting this heterogeneity in the interaction between the presence of the electric grid and land characteristics, we create a measure of incidental coverage. We anticipate that areas with higher incidental coverage exhibit higher 3G rates. Figure 3, right panel, supports this expectation, as municipalities below and above the median of incidental coverage had zero 3G Internet before 2011, after which their trends diverge in a monotonic manner.
To further support our preliminary evidence, we provide additional analysis in Table A1 in the appendix. This balance table compares the mean values of geographical and socioeconomic indicators, before the introduction of 3G in 2011, for municipalities in different quartiles of the distribution of lightning strikes and incidental coverage. This analysis allows us to explore the potential correlation between our instruments and the predetermined characteristics of the municipalities. Regarding lightning strikes, the results are particularly reassuring. In each of the examined quartiles, the two groups of municipalities exhibit similar socioeconomic indicators, suggesting comparability in terms of pre-existing characteristics. While there are slight differences in some geographical characteristics, these variations are not substantial and are difficult to link with a change in the trend in access to credit unrelated to the introduction of the 3G. As for incidental coverage, the results are similar. Apart from some correlations with socioeconomic indicators around the median, there are no significant differences throughout the distribution. Although the absolute differences in mean values around the median are small, we further control for these correlations to ensure that our results are not biased, as discussed in the robustness section.
Overall, the analyses in Figure 3 and Table A1 support the validity of our instruments by demonstrating that the two groups of municipalities, divided based on lightning strikes frequency and incidental coverage, show different trends after the introduction of the 3G while they are largely comparable in terms of pre-existing geographical and socioeconomic characteristics.
Importantly, we define our IVs as the interaction between lightning strikes and incidental coverage with a dummy post-2010. This definition further mitigates potential bias arising from the non-random allocation of mobile internet across different municipalities, and helps us identify the causal effect of mobile internet on banking.
3.3 Empirical Methodology
To examine the effects of mobile Internet on access to credit, the probability of switching to banking, and bank loan characteristics, we employ a balanced panel dataset at the borrower-year level.7 Our baseline model is the following:
where i denotes the borrower, m her municipality, and t the year. The dependent variable Yimt differs depending on the specification. In the main one, which focuses on access to credit, Yimt is a binary variable that takes the value of 1 if an individual i, in municipality m, has an outstanding loan with any financial institution, or separately with a commercial bank or a MFI, at time t, and 0 otherwise. When examining the transition to banks or MFIs, Yimt is a binary variable that equals 1 when individual i, in municipality m, switches to the banking sector (or the MFI sector) at time t or before, and 0 otherwise. When focusing on loan characteristics, Yimt is a continuous variable that represents the amount of the outstanding loan or the associated interest rate. The key independent variable of interest is 3G Coveragemt, which is the standardized measure of the percentage of mobile Internet coverage in municipality m, at time t. We control for borrower fixed effects (αi) to account for unobserved time-invariant characteristics of borrowers that may be correlated with the dependent variable. Additionally, we include time-fixed effects (ϕt) to capture common time-varying shocks affecting all borrowers simultaneously, such as changes in economic conditions. We estimate Equation (1) using a linear probability model (LPM) and compute robust standard errors clustered at the municipality level to address potential heteroscedasticity and correlated errors within the same municipality.8
By estimating this model, we examine the relationship between 3G internet coverage and access to credit, switching behavior, and loan characteristics while controlling for individual and time-specific factors.9
To address endogeneity concerns, we employ a two-stage least squares (2SLS) methodology. The first stage, outlined earlier, exploits the relationships between mobile internet coverage and the two instruments: lightning frequency and incidental coverage. The 2SLS equations are as follows:
where 3G Coveragemt is instrumented by
Equation (3) represents the second stage, where the dependent variable (Yimt) is regressed on the predicted values of 3G Coveragemt (denoted as 3G
The identification assumption underlying our model is that any correlation between lightning strikes and incidental coverage with municipality characteristics remained unchanged at the time of the introduction of the 3G technology. Indeed, we identify the effect of the change in the impact of lightning strikes and incidental coverage on the outcomes of interest, assuming that any changes in this impact occurred solely due to the update in mobile internet technology.
Table 2 presents the first-stage estimates as specified in Equation (2). Columns 1 to 3 correspond to the sample of Rwandan municipalities, with one observation for each municipality-year. Column 1 uses lightning strikes as the sole instrument for 3G coverage. Column 2 employs incidental coverage as instrument. Column 3 includes both instruments simultaneously. Columns 4 to 6 reconsider the sample of all borrowers included in the credit registry. The coefficients for lightning strikes are negative and statistically significant, while the coefficients for incidental coverage are positive and statistically significant, aligning with our hypothesis. Furthermore, the F-statistics are generally high, surpassing typical threshold values, indicating instrument relevance.
4 Results
This section outlines the main findings of our study, emphasizing the impact of mobile internet on access to bank credit, the transition of borrowers from microfinance to formal banking, and bank loan characteristics. Additionally, we present evidence on the mechanisms behind our findings, showing the complementarity between technology and law in creating a collateral system. Finally, we perform robustness checks to validate the reliability of our results. To estimate these effects, we follow the empirical methodology outlined in Section 3.
4.1 Access to Bank Credit
We investigate the impact of mobile internet on borrowers’ access to the bank credit by constructing a balanced panel dataset covering the period from 2008 to 2015.12 For each borrower-year pair, we identify whether the borrower has an outstanding loan with any financial institution, and separately, whether the loan is obtained from a commercial bank or a microfinance institution. We define three dummy variables: Probability (Any Loan)imt, indicating whether borrower i, in municipality m, has an outstanding loan in year t; Probability (Bank Loan)imt, indicating whether the loan is from a bank; and Probability (MFI Loan)imt, indicating whether the loan is from a MFI. We use these as the dependent variables in the baseline OLS regression of Equation (1).
The results, shown in Table 3, suggest a mild positive relationship between mobile internet and the probability of getting a loan, although the coefficient is statistically indistinguishable from zero. When we decompose the results between loans granted by commercial banks and microfinance institutions, we observe considerably different patterns. Column 2 of Table 3, which refers to Probability (Bank Loan)imt, shows a positive and statistically significant coefficient, indicating that higher 3G coverage in the municipality is associated with a higher probability of obtaining a loan from a commercial bank. On the other hand, column 3, which refers to Probability (MFI Loan)imt, shows a negative and statistically significant coefficient, suggesting that mobile internet is negatively associated with the likelihood of accessing loans from MFIs.
To address endogeneity concerns and claim for causality, we replicate our analysis using the 2SLS methodology in equation (3). The results, reported in columns 4 to 6 of Table 3, confirm those from the OLS regressions. The effect of mobile internet on the probability of getting a loan remains statistically indistinguishable from zero. However, when examining the decomposition, we find a significant positive effect on the probability of obtaining a loan from a commercial bank and a significant negative effect on the probability of obtaining a loan from a microfinance institution. Thecoefficients indicate that a one standard deviation increase in 3G coverage is associated with a 3.4 percentage point increase in the probability of having a loan from a bank (column 5) and a 3.6 percentage point decrease in the probability of having a loan from a MFI (column 6).13 These results imply that the impact of mobile internet on access to credit is heterogeneous: it positively affects access to formal banking, while negatively affecting access to microfinance.
Our findings corroborate the idea that mobile internet is a an important channel which contributes to financial graduation and the move towards commercial banks. On the extensive margin, potential borrowers endowed with the new internet technology prefer commercial banks to MFIs. On the intensive margin, incumbent borrowers transition to formal banking. This reallocation story is further supported by direct evidence presented in Section 4.2.
In low- and middle-income countries, including those in Africa, efforts to establish a developed financial system that promotes long-term credit and economic growth have often fallen short. Numerous regional and national policies have been implemented over the past four decades to facilitate this development but with limited success (Beck and Cull, 2013). The fact that new technologies, such as mobile internet, facilitate the transition from an informal microcredit system to a formal and well-structured banking system is a powerful result. Previous studies have shown the importance of microfinance as a first stage for borrowers, who then seek banking credit (Banerjee et al., 2015; Agarwal et al., 2023). Other studies have highlighted the transformative role of credit bureaus in underdeveloped countries, enabling borrowers to build credit history and become transparent to banks (Pagano and Jappelli, 1993; Padilla and Pagano, 1997). Some research has even demonstrated the complementary actions between microfinance and credit bureaus (De Janvry et al., 2010; Agarwal et al., 2023). What this paper adds to the literature is the identification of another potential factor, internet technologies, which smooths the transition from an informal microcredit sector to a formal and well-developed banking sector.
4.2 Direct Evidence on the Transition to Banks
We examine the transition of borrowers from microfinance institutions to commercial banks, by introducing in our setting a dependent variable (Probability (of Switching)imt), which measures the probability that a borrower, with her initial credit account in a MFI (bank), switching to a bank (MFI) at a later stage.
The analysis proceeds as follows. We identify four types of borrowers: 1) those who only have relationships with banks throughout the period analyzed, 2) those who only have relationships with MFIs, 3) those who initially have a relationship with a MFI and later switch to the banking sector, and 4) those who initially have a relationship with a bank and later access a MFI. The sample is then divided into two groups: “First MFI” consisting of borrowers whose initial relationship was with a MFI; and “First bank” consisting of borrowers whose initial relationship was with a commercial bank. Separately for each group, we estimate OLS and 2SLS regressions with Probability (of Switching)imt as the dependent variable, using equations (1) and (3).
The results are presented in Table 4. Columns 1 and 2 focus on the probability of switching to banks for borrowers whose initial relationship was with a MFI, using OLS and 2SLS respectively. Estimates from column 2 indicate that a one standard deviation increase in 3G mobile coverage is associated with a 1.1 percentage point increase in the probability of switching to banks. Columns 3 and 4 focus on the probability of switching to MFIs for borrowers who started with a banking relationship. Contrary to the previous findings, estimates from column 4 show that a one standard deviation increase in 3G mobile coverage is associated with a 1.1 percentage point decrease in the probability of switching to MFIs.
These results are consistent with those in section 4.1 and provide direct evidence of the positive effect of mobile Internet on the intensive margin of the transition from microfinance to formal banking. When endowed with mobile internet, incumbent MFI borrowers relatively increase their probability of moving to commercial banks.
Given the focus of the paper on the banking sector, all subsequent analyses are dedicated to examining the impact of mobile internet on banks and bank loans.
4.3 Heterogeneity in Access to Bank Credit
We investigate whether the effect of mobile internet on accessing a commercial bank loan is heterogeneous along different dimensions. We modify the baseline specifications in Equations (1) and (3) considering time, socio-economic, and borrower-specific factors that may influence the relationship between our main predictor and access to bank loans. We rely on a dynamic specification to account for time heterogeneity, while we use splitting samples techniques to highlight potential differences coming from socio-economic and borrower-specific attributes. Our results provide insights into which groups benefit the most from the availability of mobile internet and how different factors shape this relationship.
We use an event study to investigate the dynamics of the effect. By focusing on the year of the introduction of 3G in each municipality, we observe the evolution over time of the effect of mobile internet by estimating the following model:
where K indicates the relative year from the introduction of the 3G (Kmt = t — year 3Gm), βk are the coefficients associated with each (relative) year, and the other variables are defined as in equation (1). We set the year before the arrival of the 3G as the omitted category.
Figure A1 shows the event study associated with the OLS regression. The graph demonstrates the lagged nature of the effect, indicating that the impact of mobile internet on bank loans takes time to materialize. The effect becomes significant after two years following the introduction of 3G and it is notable in magnitude. Importantly, the analysis allows us to highlight the absence of pre-trends, suggesting that there were no systematic differences in bank loan access trends between municipalities with and without 3G prior to its introduction, reinforcing the validity of our identification strategy. Overall, the event-study analysis provides compelling evidence of the lagged and significant effect of mobile internet on accessing a commercial bank loan.
In order to examine other forms of heterogeneity, we conduct additional tests based on the socio-economic characteristics of the municipalities in our sample. First, we split the sample into rural and urban municipalities based on the share of the urban population. Second, we divide the municipalities into poorer and wealthier using data from Household Surveys (EICV). Finally, we distinguish municipalities with a physical bank branch from those without one. The results of these tests are presented in Table 5, panel A.14 We observe that the effect of 3G coverage appears to be larger in urban and wealthier areas, and where a physical bank branch is present. Although none of these differences are statistically significant, our findings suggest that the effect of mobile internet on accessing a commercial bank loan may be more pronounced in more developed regions, being negligible in extremely rural and the poorest areas of the country. Overall, these results highlight the importance of considering local economic conditions when analyzing the effects of internet technologies on banking.
To further investigate the potential heterogeneous effects of mobile internet, we leverage the individual characteristics of the borrowers in our sample. We divide the sample based on the gender, marital status (single versus married), and age (young versus adult) of the borrower. The results of these tests are summarized in Table 5, panel B. We find that the effect of 3G coverage on the probability of obtaining a loan from a commercial bank is homogeneous across different borrower characteristics. Although there may be slight variations in the magnitudes, these differences are not statistically significant. Whether borrowers are female or male, single or married, young or adult, the effect on their likelihood of obtaining a loan from a commercial bank remains relatively similar. This uniform effect implies that the adoption of mobile internet has broad positive implications, regardless of individual borrower characteristics.
The latter has important implications for policymakers, which may create an enabling environment for individuals to access and utilize formal banking services by promoting the development of mobile infrastructure and improving internet connectivity.
4.4 Bank Loan Characteristics
By investigating the relationship between 3G coverage and loan characteristics, our analysis aims to shed light on how the availability of mobile internet influences the nature and extent of bank lending activities. We exploit the granularity of our data and present results at the individual borrower-bank-year level, focusing on borrowers who have accessed bank loans. In this way, we examine the impact of mobile internet along the intensive margin of the credit relationship.15
Our primary findings are presented in Table 6, columns 1 and 2 and show that a one standard deviation increase in 3G coverage is associated with an increase in outstanding loan amount by 6.5%, which indicates that borrowers who have access to 3G tend to have larger outstanding loan balances. At the same time, we find no evidence of an effect on interest rates, i.e. borrowers with access to mobile internet show larger outstanding loan amounts without experiencing adverse effects on loan prices. These findings point to the beneficial role of mobile internet in enhancing borrowing opportunities and loan terms for individuals accessing commercial bank loans, and are corroborated by aggregate-level results in Table B1 in the appendix.
As a second step, we focus on borrowers who initially have a loan with a MFI, exploring the heterogeneous effect of mobile internet on loan conditions for those who switched to a commercial bank.16 The model estimated is the following:
where Yimt measures either the amount of loan outstanding or the average interest rate for borrower i, in municipality m, in year t. The main predictors are: i) the standardized 3G coverage
The results, presented in Table 6, columns 3 and 4, provide insights into how mobile internet affects the loan characteristics of borrowers who switch from MFIs to commercial banks. Different from the borrowers keeping their relationships with MFIs, for whom access to the 3G internet has no effect, switchers experience a relative increase in loan outstanding and a significant decrease in the average interest rate.17 These findings highlight the role of mobile internet in facilitating and enhancing the loan conditions for switchers.
The results in Table 6 provide compelling evidence that mobile internet influences bank loan characteristics. Columns 3 and 4, which specifically focus on borrowers who switch to banks, indicate that mobile internet particularly contributes to better credit terms and improved financial opportunities for switchers. Taken together, these findings have relevant implications for policymakers, financial institutions, and individuals seeking to leverage the benefits of new internet technologies.
4.5 The Complementarity between Technology and Law
At the core of the paper lies the hypothesis that technology reduces the costs associated with accessing and utilizing property rights, enabling individuals to obtain land titles and leverage them as collateral for loans. This synergy between technology and law allows individuals to overcome traditional financial barriers that hinder their access to banks, thereby making microfinance institutions and informal loans the primary options for most individuals.
This section aims to test this hypothesis. We partner with the Rwanda Land Management and Use Authority and gain access to comprehensive information on land certificates, transactions, and mortgaged lands at the municipality level.
4.5.1 The Collateral Channel
One of the milestones of the Rwandan economy is the land, which is the main source of income for most of the population. In the aftermath of the civil conflict, the large-scale migration of returnees had a dramatic impact on land ownership, escalating tensions and land conflicts, making land reform a crucial condition to avoid social tensions, ensure social stability, and improve land use management and investments. With this background, in 2009 the government of Rwanda launched the Land Tenure Regularization program. This reform aimed to regulate the ownership and control of the lands, enhance land utilization, and promote its efficient management and administration (Abbott and Mugisha, 2015).
Within the LTR initiative, data on land parcels were digitized to facilitate the issuance of ownership certificates. By the year 2012, the main database had accumulated information on more than 10.4 million land parcels. In this way, the LTR program introduced a land tenure system based on the digital registry (World Bank, 2020), which played a pivotal role in favoring access to formal credit. In fact, all banks in Rwanda were connected to the digital registry, which made it easier the tracking and verification of land ownership—a crucial step in allowing the third-party validation of titles as collateral.
Mobile internet played a crucial role in informing people and disseminating the reform, especially its procedures, purposes, and technical aspects. Government agencies, officials, and institutions extensively utilized social media platforms, online newspapers, and micro-blogging sites to connect with the public and provide timely updates on the program’s progress.18 Furthermore, the widespread use of mobile phones in Rwanda made disseminating information through the mobile internet particularly effective. In this way, the LTR program reached a larger audience and engaged with individuals directly, such as through social networking forums.
To provide evidence that the collateral channel explains the effects of mobile internet on banking, we provide five pieces of evidence. First, we present statistics on the speed of registration of land transactions, and show the relevance of the LTR in reducing transaction costs. Second, we show the concurrent positive relationship between 3G internet coverage and the number of land title transactions. Third, we examine the association between 3G internet coverage and the amount of mortgaged land parcels. Fourth, we analyze the relationship between 3G internet coverage and bank credit collateral. Finally, we quantify the extent to which the effect of mobile internet on banking outcomes is mediated through the collateral channel.
Figure 4 illustrates the average number of days required to register a land transaction between 2004 and 2019.19 The solid red line represents Rwanda, while the dashed green line the average across Sub-Saharan African countries. Prior to the reform, Rwanda ranked among the poorest performers in Africa, with an average registration time exceeding 350 days, compared to the Sub-Saharan average of approximately 100 days. However, following the introduction of the reform, there has been a significant improvement, with the registration time plummeting from over 300 days to less than 40 days in under five years. This transformation propelled Rwanda from being one of the poorest performers to a top performer, leading the continent in terms of registration speed. This technical advancement translated into tangible outcomes, largely due to the concurrent rollout of mobile internet, which played a crucial role in disseminating information about the reform, including its procedures, objectives, and technical aspects, thereby facilitating its implementation. Table 7 directly focuses on the relationship between 3G coverage and land transactions, using data at the municipality level. The empirical methodology is the same as in Equation (3), with data aggregated at the municipality level. The only difference is that the period of analysis ends in 2013, which includes the hot phase of the LTR program.20 Our dependent variables are: 1) 50% Benchmark, a dummy variable indicating that at least 50% of the transactions have been made in the first year of the program; and 2) Asinh (Number of Transactions), the inverse hyperbolic sine transformation of the cumulative of the total number of transactions in municipality m and year t, scaled by the population in the municipality. Estimates in Table 7 show a positive relationship between 3G Internet coverage and our proxies of the availability of land certificates, suggesting a role for mobile internet in promoting the LTR program and facilitating the marketing of land titles. Specifically, the 2SLS results in columns 3 and 4 show that a one standard deviation increase in 3G coverage is associated with a 15 percentage point higher probability of disbursing at least 50% of the titles in the first year of the reform, and a 13% increase in the total number of transactions.21
Next, in Table 8, columns 1 and 2, we study the relationship between 3G internet coverage and the amount of mortgaged lands. Our dependent variable is the share of mortgaged parcels over the total amount of parcels. Since these data are only available in cross-section, our results are suggestive of the relationship between mobile internet and mortgaged lands.22 In column 1, we report the OLS cross-sectional results, where the share of mortgaged lands is regressed on the level of 3G coverage in 2011. In column 2 we report the associated 2SLS estimates. The results show a positive relationship between mobile internet and mortgaged parcels. Specifically, a one standard deviation increase in 3G coverage is associated with a 1.7 percentage point increase in the share of land parcels used as collateral.
Table 8, columns 3 to 6, provides further insights into the relationship between mobile internet and access to collateralized loans and bank mortgages. The analysis is based on a balanced panel dataset from the credit registry, and the methodology follows Equation (3). Our dependent variables are: 1) Probability (Bank Collateral), a dummy variable that indicates whether the bank loan is backed by collateral; and 2) Probability (Bank Mortgage), a dummy variable indicating whether the loan is a mortgage issued by a commercial bank. The results confirm those in the previous columns. A one standard deviation increase in 3G coverage is associated with a 0.6 percentage point increase in the probability of a bank loan being backed by collateral (column 5) and a 0.8 percentage point increase in the probability of the loan being a mortgage issued by a commercial bank (column 6). This suggests that individuals in areas with better access to mobile internet have a higher probability of benefiting from collateralized loans.
Our findings, taken together, provide robust evidence on the role of mobile internet in promoting land reform and its impact on accessing formal credit through the collateral channel.
To conclude and refne our analysis, we offer a Sobel-Goodman mediation test to quantify how much of the effect of mobile internet on banking is mediated through the effect of the former on the LTR (Doerr et al., 2022).23 We report estimates from the Sobel-Goodman test in Table 9. Consistently with our story, we find that approximately 35% of the probability of receiving a bank loan is due to an indirect effect facilitated by the acquisition of the collateral,24 whereas 65% is directly attributed to mobile internet. The “direct effect” of mobile Internet is a residual effect and catches many potential channels through which the internet may make individuals seek bank credit: better job opportunities (Hjort and Poulsen, 2019), higher income (Bahia et al., 2023; Calderone et al., 2018), and lower information frictions (Gupta et al., 2023). Then, when repeating our estimation using as dependent variable the probability of receiving a collateralized loan, we find that the direct effect of mobile internet decreases to 20% and is borderline significant, while the indirect effect mediated by the land titles accounts for 80% of the total impact. This quantification points to the pivotal role of collateral in securing a mortgage and steering individuals from MFIs, which lack this product, towards commercial banks.
Overall, our results highlight the interplay between mobile internet, collateral, and access to formal credit, and underscore the importance of considering the links coming from different markets when evaluating the effects of a general-purpose technology such as mobile internet.
4.6 Robustness
We conduct a series of robustness checks to gauge the sensitivity of our main results to different specifications.
Firstly, we replicate our analysis on banking by aggregating data at the bank-municipality-year level. We estimate the following model:
where b denotes the commercial bank, m the municipality, and t the year. The dependent variable Ybmt captures different loan characteristics, including the number of loan relationships, the number of switching borrowers from microfinance institutions, the total amount of outstanding loans, and the average interest rate. Bank-municipality fixed effects (αbm) account for unobserved time-invariant characteristics that may be specific to each bank-municipality pair, while bank-year fixed effects (αbt) capture common time-varying shocks that affect the bank in a particular year. Estimates, reported in Table B1, confirm those in the main text. A one standard deviation increase in 3G coverage is associated with a 13% increase in the number of loans granted, a similar increase in the number of borrowers switching from the microfinance sector, a 33% increase in total loan outstanding, and no significant effect on average interest rates.
Then, we make several modifications to the main 2SLS specification reported in Table 3.
Our original dataset is made of 190,138 individual borrowers. Since we lack information on the age of some of them, we decide to drop these borrowers from the main analysis.25 As a first robustness check, we test the sensitivity of our results to this choice. Estimates from the model considering the full sample are reported in Table B2 in the appendix. As we can see from the table, all the coefficients are unaffected. They keep their sign and magnitudes, as well as their statistical significance.
When studying the effect of mobile internet on access to bank credit, our main dependent variable is a dummy indicating whether the borrower has an outstanding loan with a bank (MFI). An alternative definition might refer to the first year since the borrower becomes a bank (MFI) client. We create a staggered measure of bank (MFI) loan that takes value 1 from the first year in which the borrower has an outstanding loan with the lender. Our analysis is reported in Table B3 in the appendix. Results are coherent with those in Table 3. The magnitude of the coefficient associated with bank loans is higher, whereas that associated with MFI loans is lower.
Our main predictor is the share of 3G coverage in the municipality. As a first robustness check, we substitute this variable with a dummy that takes value 1 from the first year in which the municipality is reached by mobile internet. Our analysis, which resembles a two-way fixed effects regression, is reported in Table B4 in the appendix. 3G coverage is associated with an increase in the probability of having a loan with a bank of about 10 percentage points, and a similar decrease in the probability of having a loan with a MFI. As a second robustness check, we create a predictor which weights 3G coverage for population density and total population. Results from these revised specifications are reported in Table B5 in the appendix and are in line with those in Table 3, with larger magnitudes.
In the main analysis, we adopt a version of the instruments where the latter are the interactions of the time-invariant component of Z and a dummy post-2010. As a robustness check, we replicate our analysis by using two alternative definitions of the instruments. First, we interact the time-invariant component of Z with a linear time trend. Second, we interact the baseline instruments with the linear time trend. Results are reported in Table B6 in the appendix and are qualitatively in line with those in Table 3, but larger in magnitudes.
Our 2SLS estimates use two instruments for one endogenous variable. The textbook motivation for this choice is statistical efficiency. However, there is a deeper reason based on the fact that this allows for heterogeneous treatment effects. In an influential paper, Imbens and Angrist (1994) show that estimand from a 2SLS with multiple IVs can be interpreted as a positively weighted average of LATEs for subpopulations whose treatment status is affected by the instruments (thus allowing for heterogeneous treatment effects). This result holds for any number of instruments, as long as the monotonicity condition is satisfied.26 Recently, Mogstad et al. (2021) extend the framework of Imbens and Angrist (1994) and show that these results can be generalized if a partial monotonicity condition is satisfied.27 Throughout the paper, we first show that our estimates do not suffer from over identification problems. Then, we provide direct evidence of the validity of using multiple instruments. First, we follow Mogstad et al. (2021) and find that, in our case, the null hypothesis of negative weights is rejected at conventional levels (p = 0.000), while a test of the null of positive weights does not reject and generates a high p-value of 1.0. Second, we replicate our analysis by instrumenting 3G coverage with a single instrument, i.e. lightning strikes and incidental coverage, separately. Results are reported in Table B7 in the appendix and are consistent with our main findings.
Table B8 in the appendix presents estimates where we replace our instruments, lightning strikes and incidental coverage, with predetermined geographical features that are inputs into these IVs: the altitude of the municipality and the percentage covered by wood, water, and sand. Our results confirm those presented in Table 3, showing larger magnitudes and higher F-statistics (above 100).
To conclude this series of tests on the 2SLS strategy, Table B9 presents reduced form regressions. Columns 1 and 2 display estimates when lightning strikes and incidental coverage are used as predictors. Columns 3 and 4 focus on altitude, wood, water, and sand coverage. Estimates from both IV strategies are consistent with our main specifications.
Our basic identification assumption can be violated if underlying trends affect the outcomes of interest and correlate with the 3G coverage. To control for these confounding factors, we augment our specifications with several economic and socio-demographic municipality characteristics. The group of control variables is selected taking into account table A1 and includes: 2G coverage, which controls for alternative mobile technologies; total population, urban population, and employed population, which control for demographic and economic features; the number of total bank branches, which control for bank physical infrastructures; and a dummy for the presence of a primary road, which controls for the level of connection of the municipality. For variables obtained from the 2012 Census, we use an interaction with the dummy post-2010. Table B10 reports the estimates related to this augmented specification. The coefficients keep the same sign as in the baseline regressions and remain statistically significant. Interestingly, the magnitudes in columns 1 and 2 are slightly affected by the inclusion of control variables.
In 2009, the government of Rwanda launched the Umurenge SACCOs (U-SACCOs) program, with the aim to boost up rural savings and increase financial inclusion. The program helped people to have access to credit through U-SACCOs, build credit history, and eventually access commercial banking (Agarwal et al., 2023). Since the program became effective in 2011, we test whether the coefficients associated with mobile internet may capture part of the (confounding) effects generated by the U-SACCOs expansion. We test this alternative hypothesis by augmenting our main specification with the inclusion of a dummy variable for the presence of a U-SACCO in the municipality. Estimates from these regressions are reported in Table B11 in the appendix. Our coefficients remain unaffected, indicating that the effect of mobile internet on the probability of having a bank loan is orthogonal to the nationwide U-SACCO program.
In 2008, the government of Rwanda began rolling out a national fiber optic backbone that was completed at the end of 2010, with over 3,000 kilometers of fiber, distributed to all 30 districts. This capillary terrestrial fiber optic backbone, which facilitates the transfer of high-speed data across the country, can be a confounder if associated with our main predictor. To test this hypothesis, we combined information on the national fiber-optic backbone from AfTerFibre Maps, with those of the Liquid Telecom private network. We generate a dummy variable for the presence of fixed broadband at the municipality level and augment our main specifications with this variable. Results are reported in Table B12 and are in line with those in Table 3.
As a last robustness exercise, we check the sensitivity of our results in Table 7, which are related to the collateral channel, to the presence of fixed broadband. When examining the impact of mobile internet on land transactions, the expectation is that the effect is specifically driven by the ability of mobile internet to facilitate the dissemination of information about the land reform program through channels such as social media. If this is the case, a horse-race test with fixed broadband, which represents a different mode of internet access, should not affect the main results. This is because fixed broadband is less likely to be directly linked to the promotion of the LTR through the web and social media platforms. To overcome the endogeneity of the presence of fixed broadband, we instrument this variable using the presence of post-colonial surfaced roads, in line with the works by Dalgaard et al. (2022) and Barjamovic et al. (2019). In particular, we exploit the fact that underground fiber-optic cables are mainly laid along highways and by city roads (Nyarko-Boateng et al., 2019; Redifer et al., 2020) together with the availability of post-colonial road maps from the Perry-Castañeda Library Map Collection of the University of Texas.28 Results from this falsification test are reported in Table B13 and provide robust evidence in favor of the collateral channel as advocated in Section 4.
5 Conclusion
This paper investigates the complementarity between technology and law in alleviating financial frictions in low and middle-income countries. Leveraging a comprehensive dataset from the credit registry of Rwanda alongside a land reform, we offer compelling evidence on the impact of mobile internet on banking. Specifically, we show that increased mobile connectivity, as indicated by expanded 3G coverage, enhances access to land titles. This, in turn, enables individuals to access bank credit by utilizing land titles as collateral.
Our findings reveal that expanded 3G coverage is associated with a higher probability of obtaining a bank loan, a decrease in the likelihood of securing a microfinance loan, and no significant change in overall access to finance. Importantly, we observe a transition among borrowers from microfinance institutions to banks, driven by the transformative role of mobile internet in fostering the land reform. This transition to bank credit is facilitated by the availability of digitally-empowered land collateral.
Overall, our study provides strong evidence of the transformative power of mobile internet in promoting the development of formal financial systems. The findings highlight the important role of mobile technology in bridging information gaps, thus facilitating access to bank credit. These insights can inform policymakers and stakeholders in designing strategies to enhance financial inclusion, not only in Rwanda but also in other economies facing similar challenges.
References
Abbott, P. and R. Mugisha (2015): “Land Tenure Regularisation Programme: Progress Report for Selected Indicators,”.
Acampora, M., L. Casaburi, and J. Willis (2022): “Land rental markets: Experimental evidence from Kenya,” Tech. rep., National Bureau of Economic Research.
Access to Finance Rwanda (2020): “Finscope Rwanda Consumer Survey Report,”.
Acharya, A., M. Blackwell, and M. Sen (2016): “Explaining causal findings without bias: Detecting and assessing direct effects,” American Political Science Review, 110, 512–529.
Agarwal, S., T. Kigabo, C. Minoiu, A. Presbitero, and A. F. Silva (2023): “Serving the Underserved: Microcredit as a Pathway to Commercial Banks,” Review of Economics and Statistics, 105, 780–797.
Ali, D. A., K. Deininger, and M. Duponchel (2017): “New Ways to Assess and Enhance Land Registry Sustainability: Evidence from Rwanda,” World Development, 99, 377–394.
Badarinza, C., V. Balasubramaniam, and T. Ramadorai (2019): “The household finance landscape in emerging economies,” Annual Review of Financial Economics, 11, 109–129.
Bahia, K., P. CASTELLS, G. CRUZ, T. MASAKI, C. RODRIGUEZ-CASTELAN, and V. SAN-FELICE (2023): “Mobile Broadband, Poverty, and Labor Outcomes in Tanzania,” The World Bank Economic Review, 37, 235–256.
Banerjee, A., E. Duflo, R. Glennerster, and C. Kinnan (2015): “The miracle of microfinance? Evidence from a randomized evaluation,” American Economic Journal: Applied economics, 7, 22–53.
Barjamovic, G., T. CHANEY, K. CO§AR, and A. HORTACSU (2019): “Trade, merchants, and the lost cities of the bronze age,” The Quarterly Journal of Economics, 134, 1455–1503.
Beck, T. and R. Cull (2013): “Banking in Africa,” Available at SSRN 2348417.
Besley, T. and M. Ghatak (2010): “Property rights and economic development,” in Handbook of Development Economics, Elsevier, Vol. 5, 4525–4595.
Besley, T. and T. Persson (2011): Pillars of Prosperity: The Political Economics of Development Clusters, Princeton University Press.
Besley, T. J., K. B. Burchardi, M. Ghatak, et al. (2012): “Incentives and the De Soto Effect,” The Quarterly Journal of Economics, 127, 237–282.
Bircan, C. and R. De Haas (2020): “The limits of lending? Banks and technology adoption across Russia,” The Review of Financial Studies, 33, 536–609.
Bjorkegren, D. (2019): “The adoption of network goods: Evidence from the spread of mobile phones in Rwanda,” The Review of Economic Studies, 86, 1033–1060.
Bos, M., E. Breza, and A. Liberman (2018): “The labor market effects of credit market information,” The Review of Financial Studies, 31, 2005–2037.
Braghieri, L., R. Levy, and A. Makarin (2022): “Social media and mental health,” American Economic Review, 112, 3660–3693.
Breza, E. and C. Kinnan (2021): “Measuring the equilibrium impacts of credit: Evidence from the Indian microfinance crisis,” The Quarterly Journal of Economics, 136, 1447–1497.
Caldarola, B., M. Grazzi, M. Occelli, and M. Sanfilippo (2023): “Mobile internet, skills and structural transformation in Rwanda,” Research Policy, 52, 104871.
Calderone, M., N. Fiala, F. Mulaj, S. Sadhu, and L. Sarr (2018): “Financial education and savings behavior: Evidence from a randomized experiment among low-income clients of branchless banking in India,” Economic Development and Cultural Change, 66, 793–825.
Core, F. and F. De Marco (2021): “Public guarantees for small businesses in Italy during COVID-19,”.
Dalgaard, C.-J., N. Kaarsen, O. Olsson, and P. Selaya (2022): “Roman roads to prosperity: Persistence and non-persistence of public goods provision,” Journal of Comparatrive Economics, 50, 896–916.
D’Andrea, A., M. Pelosi, and E. Sette (2023): “When Broadband Comes to Banks: Credit Supply, Market Structure, and Information Acquisition,” Market Structure, and Information Acquisition.
De Janvry, A., C. McIntosh, and E. Sadoulet (2010): “The supply-and demand-side impacts of credit market information,” Journal of Development Economics, 93, 173–188.
De Mel, S., D. McKenzie, C. Woodruff, et al. (2009): “Returns to Capital in Microenter-prises: Evidence from a Field Experiment,” The Quarterly Journal of Economics, 124, 423–423.
De Soto, H. (2000): The mystery of capital: why capitalism triumphs in the West and fails everywhere else, Basic Books/New York.
Deininger, K. and A. Goyal (2012): “Going digital: Credit effects of land registry computerization in India,” Journal of Development Economics, 99, 236–243.
Djankov, S., E. L. Glaeser, V. Perotti, and A. Shleifer (2020): “Measuring property rights institutions,” Working Paper.
Djankov, S., C. McLiesh, and A. Shleifer (2007): “Private credit in 129 countries,” Journal of Financial Economics, 84, 299–329.
DOERR, S., S. GlSSLER, J.-L. PEYDRO, and H.-J. VOTH (2022): “Financial crises and political radicalization: How failing banks paved Hitler’s path to power,” The Journal of Finance, 77, 3339–3372.
D’Andrea, A. and N. Limodio (2024): “High-Speed Internet, Financial Technology, and Banking,” Management Science, 70, 773–798.
Enikolopov, R., A. Makarin, and M. Petrova (2020): “Social media and protest participation: Evidence from Russia,” Econometrica, 88, 1479–1514.
Fazio, D. and T. Silva (2021): “Housing Collateral Reform and Economic Reallocation,” Working Paper.
Fujiwara, T., K. MtJLLER, and C. SCHWARZ (2023): “The effect of social media on elections: Evidence from the United States,” Journal of the European Economic Association, in press.
Fuster, A., M. Plosser, P. Schnabl, and J. Vickery (2019): “The role of technology in mortgage lending,” The Review of Financial Studies, 32, 1854–1899.
Gao, M. and J. Huang (2020): “Informing the market: The effect of modern information technologies on information production,” The Review of Financial Studies, 33, 1367–1411.
Gavazza, A., M. Nardotto, and T. Valletti (2019): “Internet and politics: Evidence from UK local elections and local government policies,” The Review of Economic Studies, 86, 2092–2135.
GoR (2012): “Rwanda Vision 2020,”.
GSMA (2015): “Rural Coverage: Strategies for Sustainability,” http://www.gsma.com/mobilefordevelopment/wp-content/uploads/2015/07/Rural coverage strategies for sustainability.pdf.
Gupta, A., J. Ponticelli, and A. Tesei (2023): “Information Frictions and Take-up of Government Credit Programs,” Available at SSRN 4496550.
Guriev, S., N. Melnikov, and E. Zhuravskaya (2021): “3g internet and confidence in government,” The Quarterly Journal of Economics, 136, 2533–2613.
Hjort, J. and J. Poulsen (2019): “The arrival of fast internet and employment in Africa,” American Economic Review, 109, 1032–79.
Hjort, J. and L. Tian (2024): “The Economic Impact of Internet Connectivity in Developing Countries,” Working Paper.
Hvide, H. K., T. G. Meling, M. Mogstad, and O. L. Vestad (2022): “Broadband internet and the stock market investments of individual investors,” Tech. rep., National Bureau of Economic Research.
Imbens, G. W. and J. D. Angrist (1994): “Identification and estimation of local average treatment effects,” Econometrica, 467–475.
Imbens, G. W. and J. M. Wooldridge (2009): “Recent Developments in the Econometrics of Program Evaluation,” Journal of Economic Literature, 47, 5–86.
Ioannidou, V., N. Pavanini, and Y. Peng (2022): “Collateral and asymmetric information in lending markets,” Journal of Financial Economics, 144, 93–121.
Jiang, E. X., G. Y. Yu, and J. Zhang (2022): “Bank Competition amid Digital Disruption: Implications for Financial Inclusion,” Available at SSRN 4178420.
Kaboski, J. P. (2024): “Financial frictions, financial market development, and macroeconomic development,” Oxford Development Studies, 51, 397–416.
Kaboski, J. P. and R. M. Townsend (2012): “The impact of credit on village economies,” American Economic Journal: Applied Economics, 4, 98–133.
Kwan, A., C. Lin, V. Pursiainen, and M. Tai (2024): “Stress Testing Banks’ Digital Capabilities: Evidence from the COVID-19 Pandemic,” Journal of Financial and Quantitative Analysis, in press, 1–29.
La Porta, R., F. Lopez-de Silanes, A. Shleifer, and R. W. Vishny (1998): “Law and finance,” Journal of Political Economy, 106, 1113–1155.
Lin, C., C. Ma, Y. Sun, and Y. Xu (2021): “The telegraph and modern banking development, 1881-1936,” Journal of Financial Economics, 141, 730–749.
Manacorda, M. and A. Tesei (2020): “Liberation technology: Mobile phones and political mobilization in Africa,” Econometrica, 88, 533–567.
Manysheva, K. (2021): “Land Property Rights, Financial Frictions, and Resource Allocation in Developing Countries,” Tech. rep., mimeo.
Mazet-Sonilhac, C. (2021): “Information Frictions in Credit Markets,” Available at SSRN 3969990.
Mogstad, M., A. Torgovitsky, and C. R. Walters (2021): “The causal interpretation of two-stage least squares with multiple instrumental variables,” American Economic Review, 111, 3663–98.
Montero, E. (2022): “Cooperative property rights and development: Evidence from land reform in El Salvador,” Journal of Political Economy, 130, 48–93.
Nyarko-Boateng, O., A. F. Adekoya, and B. A. Weyori (2019): “Investigating QoS and Performance of Received Signal Strength Indicator in Fiber Optics Broadband Data Communication,” AJEAS. DOI, 10, 391–401.
Padilla, A. J. and M. Pagano (1997): “Endogenous communication among lenders and entrepreneurial incentives,” The Review of Financial Studies, 10, 205–236.
Pagano, M. and T. Jappelli (1993): “Information sharing in credit markets,” The Journal of Finance, 48, 1693–1718.
Pierri, N. and Y. Timmer (2022): “The importance of technology in banking during a crisis,” Journal of Monetary Economics, 128, 88–104.
Ponticelli, J. and L. S. Alencar (2016): “Court enforcement, bank loans, and firm investment: evidence from a bankruptcy reform in Brazil,” The Quarterly Journal of Economics, 131, 1365–1413.
Redifer, L., E. Alper, N. Meads, T. Gursoy, M. Newiak, A. H. Thomas, and S. Kwalingana (2020): “The Development Path Less Traveled: The Experience of Rwanda,” Departmental Papers, 2020.
Schreiber, L. (2017): “Securing Land Rights: Making Land Titling Work in Rwanda: 2012–2017,”.
UN-OHRLLS (2017): “Leveraging Investments in Broadband for National Development: The Case of Rwanda and Senegal.”.
World Bank (2018): “Rwanda Afordable Housing Finance Project, Combined Project Information Documents / Integrated Safeguards Datasheet,”.
World Bank (2020): “Enhancing Government Effectiveness and Transparency: The Fight Against Corruption,”.
Yanagizawa-Drott, D. (2014): “Propaganda and conflict: Evidence from the Rwandan genocide,” The Quarterly Journal of Economics, 129, 1947–1994.
Figures
3G coverage in Rwanda
Citation: IMF Working Papers 2024, 070; 10.5089/9798400270734.001.A001
Notes: This figure plots the diffusion of the 3G mobile internet by municipality in Rwanda (percentage of coverage). The panel on the left refers to 2011, the first year of the introduction of 3G. The panel on the right refers to 2015, the last year for which we have data on 3G coverage. Source: GSMA.Cross-sectional variation mobile internet: Lightning frequency and incidental coverage
Citation: IMF Working Papers 2024, 070; 10.5089/9798400270734.001.A001
Notes: This figure shows the average frequency of lightning strikes and incidental coverage, by municipality in Rwanda. On the left panel, there is lightning frequency, a time-invariant indicator computed as the yearly average of strikes in the municipality, during the period 1998–2013. On the right panel, is incidental coverage (Björkegren, 2019), a time-invariant measure of the fctitious coverage that would have resulted had the operator built towers for mobile internet along the full network of power lines. Sources: GHRC and Björkegren (2019).3G coverage, lightning frequency, and incidental coverage
Citation: IMF Working Papers 2024, 070; 10.5089/9798400270734.001.A001
Notes: This figure provides visual evidence in support of the relevance of our instruments. On the left, we focus on lightning strikes. On the y-axis is a measure of 3G mobile coverage. On the x-axis are years. The solid blue line refers to municipalities with lower than median lightning frequency. The solid red line refers to municipalities with above median lightning frequency. On the right, we focus on incidental coverage. On the y-axis is a measure of 3G mobile coverage. On the x-axis are years. The solid blue line refers to municipalities with higher than median incidental coverage. The solid red line refers to municipalities with lower than median incidental coverage.Registration time for land titles
Citation: IMF Working Papers 2024, 070; 10.5089/9798400270734.001.A001
Notes: This figure plots the average number of days to register a land transaction between 2004 and 2019. The red solid line represents Rwanda, while the green dashed line shows the average across Sub-Saharan Africa countries. The vertical lines sign the years of the LT reform. Source: World Bank Doing Business.Tables
Summary statistics
Summary statistics
| n | mean | sd | p50 | min | max | |
|---|---|---|---|---|---|---|
| PANEL A: Access to bank credit | ||||||
| Probability of Any Loan | 909168 | 0.28 | 0.45 | 0.00 | 0.00 | 1.00 |
| Probability of Bank Loan | 909168 | 0.14 | 0.34 | 0.00 | 0.00 | 1.00 |
| Probability of MFI Loan | 909168 | 0.16 | 0.37 | 0.00 | 0.00 | 1.00 |
| Switching borrowers | 909168 | 0.09 | 0.29 | 0.00 | 0.00 | 1.00 |
| PANEL B: Bank loan characteristics | ||||||
| Outstanding loan | 124,806 | 2011.88 | 3665.80 | 693.11 | 16.67 | 27859.83 |
| Principal loan | 125,507 | 2820.34 | 4706.01 | 1062.00 | 30.00 | 34000 |
| Interest rate | 124,388 | 19.20 | 5.73 | 19.00 | 6.14 | 70.10 |
| Collateral | 128,639 | 0.12 | 0.32 | 0.00 | 0.00 | 1.00 |
| PANEL C: 3G coverage | ||||||
| 3G coverage | 2696 | 0.12 | 0.28 | 0.00 | 0.00 | 1.00 |
| Std 3G coverage | 2696 | 0.03 | 1.02 | -0.40 | -0.40 | 3.24 |
| PANEL D: Instrumental variables | ||||||
| Lightning frequency | 2696 | 0.04 | 0.02 | 0.03 | 0.02 | 0.10 |
| Std Lightning | 2696 | 0.03 | 0.83 | 0.00 | -1.24 | 4.10 |
| Incidental coverage | 2696 | 10.66 | 12.20 | 9.54 | 0.00 | 135.43 |
| Std Incidental | 2696 | 0.01 | 0.83 | 0.00 | -0.91 | 10.81 |
Summary statistics
| n | mean | sd | p50 | min | max | |
|---|---|---|---|---|---|---|
| PANEL A: Access to bank credit | ||||||
| Probability of Any Loan | 909168 | 0.28 | 0.45 | 0.00 | 0.00 | 1.00 |
| Probability of Bank Loan | 909168 | 0.14 | 0.34 | 0.00 | 0.00 | 1.00 |
| Probability of MFI Loan | 909168 | 0.16 | 0.37 | 0.00 | 0.00 | 1.00 |
| Switching borrowers | 909168 | 0.09 | 0.29 | 0.00 | 0.00 | 1.00 |
| PANEL B: Bank loan characteristics | ||||||
| Outstanding loan | 124,806 | 2011.88 | 3665.80 | 693.11 | 16.67 | 27859.83 |
| Principal loan | 125,507 | 2820.34 | 4706.01 | 1062.00 | 30.00 | 34000 |
| Interest rate | 124,388 | 19.20 | 5.73 | 19.00 | 6.14 | 70.10 |
| Collateral | 128,639 | 0.12 | 0.32 | 0.00 | 0.00 | 1.00 |
| PANEL C: 3G coverage | ||||||
| 3G coverage | 2696 | 0.12 | 0.28 | 0.00 | 0.00 | 1.00 |
| Std 3G coverage | 2696 | 0.03 | 1.02 | -0.40 | -0.40 | 3.24 |
| PANEL D: Instrumental variables | ||||||
| Lightning frequency | 2696 | 0.04 | 0.02 | 0.03 | 0.02 | 0.10 |
| Std Lightning | 2696 | 0.03 | 0.83 | 0.00 | -1.24 | 4.10 |
| Incidental coverage | 2696 | 10.66 | 12.20 | 9.54 | 0.00 | 135.43 |
| Std Incidental | 2696 | 0.01 | 0.83 | 0.00 | -0.91 | 10.81 |
First stage regressions
First stage regressions
| (1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
| All municipalities | Credit Registry sample | |||||
| Standardized 3G coverage | ||||||
| Std Lightning | -0.156*** (0.048) |
-0.141*** (0.046) |
-0.341*** (0.079) |
-0.330*** (0.079) |
||
| Std Incidental | 0.250*** (0.052) |
0.240*** (0.050) |
0.231*** (0.067) |
0.207*** (0.064) |
||
| Municipality FE | Yes | Yes | Yes | No | No | No |
| Borrower FE | No | No | No | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| F-statistic | 10.64 | 23.17 | 17.74 | 18.49 | 11.81 | 16.22 |
| Obs. | 2696 | 2696 | 2696 | 909165 | 909165 | 909165 |
| Adj. R sq. | 0.540 | 0.551 | 0.557 | 0.607 | 0.593 | 0.613 |
| Mean Dep. Var. | 0.032 | 0.032 | 0.032 | 0.276 | 0.276 | 0.276 |
| S.D. Dep. Var. | 1.024 | 1.024 | 1.024 | 1.302 | 1.302 | 1.302 |
First stage regressions
| (1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
| All municipalities | Credit Registry sample | |||||
| Standardized 3G coverage | ||||||
| Std Lightning | -0.156*** (0.048) |
-0.141*** (0.046) |
-0.341*** (0.079) |
-0.330*** (0.079) |
||
| Std Incidental | 0.250*** (0.052) |
0.240*** (0.050) |
0.231*** (0.067) |
0.207*** (0.064) |
||
| Municipality FE | Yes | Yes | Yes | No | No | No |
| Borrower FE | No | No | No | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| F-statistic | 10.64 | 23.17 | 17.74 | 18.49 | 11.81 | 16.22 |
| Obs. | 2696 | 2696 | 2696 | 909165 | 909165 | 909165 |
| Adj. R sq. | 0.540 | 0.551 | 0.557 | 0.607 | 0.593 | 0.613 |
| Mean Dep. Var. | 0.032 | 0.032 | 0.032 | 0.276 | 0.276 | 0.276 |
| S.D. Dep. Var. | 1.024 | 1.024 | 1.024 | 1.302 | 1.302 | 1.302 |
Mobile Internet and the Probability of having a loan
Mobile Internet and the Probability of having a loan
| (1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
| Probability of | ||||||
| Any Loan | Bank Loan | MFI Loan | Any Loan | Bank Loan | MFI Loan | |
| Std 3G coverage | 0.002 (0.002) |
0.018*** (0.003) |
-0.017*** (0.003) |
-0.002 (0.006) |
0.034*** (0.012) |
-0.036*** (0.012) |
| Estimation | OLS | OLS | OLS | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| SW F-statistic | 16.22 | 16.22 | 16.22 | |||
| Obs. | 909165 | 909165 | 909165 | 909165 | 909165 | 909165 |
| Adj. R sq. | 0.480 | 0.374 | 0.370 | |||
| Mean Dep. Var. | 0.284 | 0.136 | 0.164 | 0.284 | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.451 | 0.343 | 0.370 | 0.451 | 0.343 | 0.370 |
Mobile Internet and the Probability of having a loan
| (1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
| Probability of | ||||||
| Any Loan | Bank Loan | MFI Loan | Any Loan | Bank Loan | MFI Loan | |
| Std 3G coverage | 0.002 (0.002) |
0.018*** (0.003) |
-0.017*** (0.003) |
-0.002 (0.006) |
0.034*** (0.012) |
-0.036*** (0.012) |
| Estimation | OLS | OLS | OLS | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| SW F-statistic | 16.22 | 16.22 | 16.22 | |||
| Obs. | 909165 | 909165 | 909165 | 909165 | 909165 | 909165 |
| Adj. R sq. | 0.480 | 0.374 | 0.370 | |||
| Mean Dep. Var. | 0.284 | 0.136 | 0.164 | 0.284 | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.451 | 0.343 | 0.370 | 0.451 | 0.343 | 0.370 |
Mobile Internet and Transition to banks
Mobile Internet and Transition to banks
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of Switching from | ||||
| MFI to Bank | Bank to MFI | |||
| Std 3G coverage | 0.007*** (0.002) |
0.011*** (0.003) |
-0.006*** (0.001) |
-0.011** (0.005) |
| Estimation | OLS | IV | OLS | IV |
| First MFI | Yes | Yes | No | No |
| First BANK | No | No | Yes | Yes |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 16.47 | 13.89 | ||
| Obs. | 514372 | 514372 | 380974 | 380974 |
| Adj. R sq. | 0.255 | 0.296 | ||
| Mean Dep. Var. | 0.010 | 0.010 | 0.034 | 0.034 |
| S.D. Dep. Var. | 0.100 | 0.100 | 0.181 | 0.181 |
Mobile Internet and Transition to banks
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of Switching from | ||||
| MFI to Bank | Bank to MFI | |||
| Std 3G coverage | 0.007*** (0.002) |
0.011*** (0.003) |
-0.006*** (0.001) |
-0.011** (0.005) |
| Estimation | OLS | IV | OLS | IV |
| First MFI | Yes | Yes | No | No |
| First BANK | No | No | Yes | Yes |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 16.47 | 13.89 | ||
| Obs. | 514372 | 514372 | 380974 | 380974 |
| Adj. R sq. | 0.255 | 0.296 | ||
| Mean Dep. Var. | 0.010 | 0.010 | 0.034 | 0.034 |
| S.D. Dep. Var. | 0.100 | 0.100 | 0.181 | 0.181 |
Heterogeneity – Socioeconomic and Borrowers’ characteristics
Heterogeneity – Socioeconomic and Borrowers’ characteristics
| (1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
| Rural | Urban | Poorer | Wealthier | No branches | Branches | |
| Probability of Bank Loan | ||||||
| Std 3G coverage | 0.009 (0.021) |
0.032* (0.017) |
0.003 (0.032) |
0.044*** (0.011) |
0.024 (0.017) |
0.034** (0.016) |
| Estimation | IV | IV | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| SW F-statistic | 10.51 | 8.47 | 9.88 | 9.31 | 12.32 | 6.64 |
| Obs. | 482259 | 426906 | 338929 | 570236 | 397445 | 511720 |
| Mean Dep. Var. | 0.099 | 0.179 | 0.106 | 0.154 | 0.107 | 0.159 |
| S.D. Dep. Var. | 0.298 | 0.383 | 0.308 | 0.361 | 0.309 | 0.366 |
| (1) | (2) | (3) | (4) | (5) | (6) | |
| Female | Male | Single | Married | Young | Adult | |
| Std 3G coverage | 0.029** (0.012) |
0.037*** (0.013) |
0.050*** (0.017) |
0.030** (0.012) |
0.037*** (0.012) |
0.030** (0.014) |
| Estimation | IV | IV | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| SW F-statistic | 14.17 | 16.89 | 12.84 | 16.46 | 16.64 | 15.62 |
| Obs. | 324226 | 584939 | 106749 | 802416 | 466877 | 442288 |
| Mean Dep. Var. | 0.137 | 0.136 | 0.178 | 0.131 | 0.136 | 0.136 |
| S.D. Dep. Var. | 0.343 | 0.343 | 0.382 | 0.337 | 0.343 | 0.343 |
Heterogeneity – Socioeconomic and Borrowers’ characteristics
| (1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
| Rural | Urban | Poorer | Wealthier | No branches | Branches | |
| Probability of Bank Loan | ||||||
| Std 3G coverage | 0.009 (0.021) |
0.032* (0.017) |
0.003 (0.032) |
0.044*** (0.011) |
0.024 (0.017) |
0.034** (0.016) |
| Estimation | IV | IV | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| SW F-statistic | 10.51 | 8.47 | 9.88 | 9.31 | 12.32 | 6.64 |
| Obs. | 482259 | 426906 | 338929 | 570236 | 397445 | 511720 |
| Mean Dep. Var. | 0.099 | 0.179 | 0.106 | 0.154 | 0.107 | 0.159 |
| S.D. Dep. Var. | 0.298 | 0.383 | 0.308 | 0.361 | 0.309 | 0.366 |
| (1) | (2) | (3) | (4) | (5) | (6) | |
| Female | Male | Single | Married | Young | Adult | |
| Std 3G coverage | 0.029** (0.012) |
0.037*** (0.013) |
0.050*** (0.017) |
0.030** (0.012) |
0.037*** (0.012) |
0.030** (0.014) |
| Estimation | IV | IV | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| SW F-statistic | 14.17 | 16.89 | 12.84 | 16.46 | 16.64 | 15.62 |
| Obs. | 324226 | 584939 | 106749 | 802416 | 466877 | 442288 |
| Mean Dep. Var. | 0.137 | 0.136 | 0.178 | 0.131 | 0.136 | 0.136 |
| S.D. Dep. Var. | 0.343 | 0.343 | 0.382 | 0.337 | 0.343 | 0.343 |
Mobile Internet and Bank loan characteristics
Mobile Internet and Bank loan characteristics
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Volume of Loans | Lending Rate | Volume of Loans | Lending Rate | |
| Std 3G coverage | 0.063** (0.025) |
0.154 (0.154) |
0.032 (0.032) |
-0.077 (0.207) |
| Std 3G coverage × Switcher | 0.159*** (0.054) |
-2.235*** (0.660) |
||
| F-test (p-value) | 0.001 | 0.001 | ||
| Estimation | IV | IV | IV | IV |
| First MFI | No | No | Yes | Yes |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | No | No | Yes | Yes |
| Bank-year FE | Yes | Yes | No | No |
| SW F-statistic | 22.44 | 21.64 | 18.96 | 19.00 |
| Obs. | 104766 | 104423 | 110391 | 79081 |
| Mean Dep. Var. | 6.750 | 18.51 | 6.609 | 20.71 |
| S.D. Dep. Var. | 1.421 | 5.472 | 1.287 | 8.603 |
Mobile Internet and Bank loan characteristics
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Volume of Loans | Lending Rate | Volume of Loans | Lending Rate | |
| Std 3G coverage | 0.063** (0.025) |
0.154 (0.154) |
0.032 (0.032) |
-0.077 (0.207) |
| Std 3G coverage × Switcher | 0.159*** (0.054) |
-2.235*** (0.660) |
||
| F-test (p-value) | 0.001 | 0.001 | ||
| Estimation | IV | IV | IV | IV |
| First MFI | No | No | Yes | Yes |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | No | No | Yes | Yes |
| Bank-year FE | Yes | Yes | No | No |
| SW F-statistic | 22.44 | 21.64 | 18.96 | 19.00 |
| Obs. | 104766 | 104423 | 110391 | 79081 |
| Mean Dep. Var. | 6.750 | 18.51 | 6.609 | 20.71 |
| S.D. Dep. Var. | 1.421 | 5.472 | 1.287 | 8.603 |
Mobile Internet and Land transactions
Mobile Internet and Land transactions
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| 50% Benchmark | Number of Transactions | 50% Benchmark | Number of Transactions | |
| Std 3G coverage | 0.173*** (0.020) |
0.047*** (0.007) |
0.147* (0.080) |
0.128*** (0.036) |
| Estimation | OLS | OLS | IV | IV |
| Municipality FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 19.19 | 19.19 | ||
| Obs. | 2022 | 2022 | 2022 | 2022 |
| Adj. R sq. | 0.577 | 0.804 | ||
| Mean Dep. Var. | 0.122 | 0.229 | 0.122 | 0.229 |
| S.D. Dep. Var. | 0.327 | 0.316 | 0.327 | 0.316 |
Mobile Internet and Land transactions
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| 50% Benchmark | Number of Transactions | 50% Benchmark | Number of Transactions | |
| Std 3G coverage | 0.173*** (0.020) |
0.047*** (0.007) |
0.147* (0.080) |
0.128*** (0.036) |
| Estimation | OLS | OLS | IV | IV |
| Municipality FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 19.19 | 19.19 | ||
| Obs. | 2022 | 2022 | 2022 | 2022 |
| Adj. R sq. | 0.577 | 0.804 | ||
| Mean Dep. Var. | 0.122 | 0.229 | 0.122 | 0.229 |
| S.D. Dep. Var. | 0.327 | 0.316 | 0.327 | 0.316 |
Mobile Internet, Land mortgaged and Bank collateral
Mobile Internet, Land mortgaged and Bank collateral
| (1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
| Share of | Probability of | |||||
| Land Mortgaged | Bank Collateral | Bank Mortgage | Bank Collateral | Bank Mortgage | ||
| Std 3G coverage | 0.024*** (0.003) |
0.017*** (0.005) |
0.009*** (0.001) |
0.003*** (0.001) |
0.006** (0.003) |
0.008*** (0.002) |
| Estimation | OLS | IV | OLS | OLS | IV | IV |
| Borrower FE | No | No | Yes | Yes | Yes | Yes |
| Year FE | No | No | Yes | Yes | Yes | Yes |
| SW F-statistic | 18.52 | 16.22 | 16.22 | |||
| Obs. | 337 | 337 | 909165 | 909165 | 909165 | 909165 |
| Adj. R sq. | 0.528 | 0.499 | 0.512 | |||
| Mean Dep. Var. | 0.014 | 0.014 | 0.015 | 0.016 | 0.015 | 0.016 |
| S.D. Dep. Var. | 0.034 | 0.034 | 0.120 | 0.125 | 0.120 | 0.125 |
Mobile Internet, Land mortgaged and Bank collateral
| (1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
| Share of | Probability of | |||||
| Land Mortgaged | Bank Collateral | Bank Mortgage | Bank Collateral | Bank Mortgage | ||
| Std 3G coverage | 0.024*** (0.003) |
0.017*** (0.005) |
0.009*** (0.001) |
0.003*** (0.001) |
0.006** (0.003) |
0.008*** (0.002) |
| Estimation | OLS | IV | OLS | OLS | IV | IV |
| Borrower FE | No | No | Yes | Yes | Yes | Yes |
| Year FE | No | No | Yes | Yes | Yes | Yes |
| SW F-statistic | 18.52 | 16.22 | 16.22 | |||
| Obs. | 337 | 337 | 909165 | 909165 | 909165 | 909165 |
| Adj. R sq. | 0.528 | 0.499 | 0.512 | |||
| Mean Dep. Var. | 0.014 | 0.014 | 0.015 | 0.016 | 0.015 | 0.016 |
| S.D. Dep. Var. | 0.034 | 0.034 | 0.120 | 0.125 | 0.120 | 0.125 |
Sobel-Goodman mediation test
Sobel-Goodman mediation test
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | Bank Collateral | |
| Sobel | 0.013** | 0.005** |
| Aroian | 0.013** | 0.005** |
| Goodman | 0.013** | 0.005** |
| Indirect Effect (Collateral) | 0.013** | 0.005** |
| Direct Effect (Mobile Internet) | 0.021** | 0.001 |
| Total Effect | 0.034** | 0.006+ |
| Proportion Indirect | 37% | 80% |
Sobel-Goodman mediation test
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | Bank Collateral | |
| Sobel | 0.013** | 0.005** |
| Aroian | 0.013** | 0.005** |
| Goodman | 0.013** | 0.005** |
| Indirect Effect (Collateral) | 0.013** | 0.005** |
| Direct Effect (Mobile Internet) | 0.021** | 0.001 |
| Total Effect | 0.034** | 0.006+ |
| Proportion Indirect | 37% | 80% |
Additional figures and tables
The dynamic effect of mobile internet on access to bank credit
Citation: IMF Working Papers 2024, 070; 10.5089/9798400270734.001.A001
Notes: This figure shows a staggered DiD event study. The treatment group is made up of borrowers in municipalities covered by the 3G technology. The control group is made up of borrowers in municipalities not reached by the 3G technology. Year 0 corresponds to the first year in which the municipality is reached by mobile internet. On the y-axis is Prob.(bank loan), the probability that the borrower has an outstanding loan with a commercial bank. The blue solid line refers to coefficients. 90% confidence intervals are also reported.Distribution of interest rates for borrowers switching from MFIs to banks
Citation: IMF Working Papers 2024, 070; 10.5089/9798400270734.001.A001
Notes: This figure plots the density function of interest rates for the sample of switchers to banks. The blue solid line refers to the density of interest rates before the switching, i.e., when the borrower is still a client of the MFI. The red solid line refers to the density after the switch.Share of mortgaged land parcels
Citation: IMF Working Papers 2024, 070; 10.5089/9798400270734.001.A001
Notes: This figure plots the share of mortgaged land parcels, by municipality in Rwanda. Source: Land Management and Use Authority.Distribution of fiber-optic cables and post-colonial roads
Citation: IMF Working Papers 2024, 070; 10.5089/9798400270734.001.A001
Notes: The left panel of the figure plots fiber-optic lines as in 2015. The public fiber was deployed in 2010, while private fiber complemented the main lines from 2014. The right panel plots surfaced roads as in 1975, from the Perry-Castañeda Library Map Collection.Balance table: Municipalities by quartiles of lightning and incidental coverage
Balance table: Municipalities by quartiles of lightning and incidental coverage
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| 25th Percentile | 50th Percentile | 75th Percentile | ||||||||
| PANEL A: Lightning Strikes (Zlight) | ||||||||||
| Below | Above | Dif. | Below | Above | Dif. | Below | Above | Dif. | N. mun. | |
| 2G coverage | .93 | .928 | (.01) | .937 | .92 | (.08) | .924 | .942 | (-.09) | 337 |
| Mean altitude | 1.499 | 1.839 | (-1.06) | 1.582 | 1.922 | (-.91) | 1.673 | 1.988 | (-.69) | 336 |
| Wood | .04 | 3.453 | (-.26) | .024 | 5.175 | (-.32) | .683 | 8.349 | (-.37) | 336 |
| Water | 3.008 | 1.042 | (.26) | 1.629 | 1.437 | (.03) | 1.866 | .535 | (.22) | 336 |
| Sand | 0 | .001 | (-.04) | .001 | 0 | (.08) | .001 | 0 | (.07) | 336 |
| Main Power line | .226 | .246 | (-.03) | .232 | .25 | (-.03) | .222 | .298 | (-.12) | 336 |
| Hospital | .143 | .115 | (.06) | .131 | .113 | (.04) | .119 | .131 | (-.03) | 336 |
| ln Population | 10.175 | 10.03 | (.31) | 10.117 | 10.016 | (.22) | 10.057 | 10.095 | (-.09) | 337 |
| ln Urban | 3.583 | 2.644 | (.15) | 3.622 | 2.138 | (.24) | 2.736 | 3.298 | (-.09) | 337 |
| Unemployment | .036 | .031 | (.13) | .037 | .028 | (.21) | .033 | .031 | (.06) | 337 |
| Night lights | .404 | .700 | (-.12) | .763 | .49 | (.09) | .674 | .484 | (.07) | 337 |
| Poverty Index | 1.049 | .984 | (.16) | .996 | 1.004 | (-.02) | .997 | 1.008 | (-.03) | 337 |
| ln Conflicts | .145 | .435 | (-.31) | .242 | .482 | (-.23) | .318 | .494 | (-.16) | 337 |
| Bank branches | .762 | .743 | (.01) | .863 | .633 | (.10) | .75 | .741 | (0.00) | 337 |
| PANEL B: Incidental Coverage (Zincid) | ||||||||||
| Above | Below | Dif. | Above | Below | Dif. | Above | Below | Dif. | N. mun. | |
| 2G coverage | .956 | .845 | (.49) | .969 | .888 | (.41) | .974 | .913 | (.36) | 337 |
| Mean altitude | 1.734 | 1.806 | (-.16) | 1.693 | 1.810 | (-.27) | 1.669 | 1.779 | (-.23) | 336 |
| Wood | .604 | 8.586 | (-.38) | .447 | 4.726 | (-.27) | .083 | 3.425 | -(.25) | 336 |
| Water | 1.756 | .866 | (.14) | 2.197 | .878 | (.18) | 3.079 | 1.026 | (.24) | 336 |
| Sand | .001 | 0 | (.03) | .001 | 0 | (.07) | 0 | .001 | (-.07) | 336 |
| Main Power line | .274 | .143 | (.23) | .287 | .195 | (.15) | .229 | .245 | (-.03) | 336 |
| Hospital | .139 | .071 | (.16) | .15 | .095 | (.12) | .084 | .134 | (-.11) | 336 |
| ln Population | 10.091 | 9.993 | (.23) | 10.111 | 10.022 | (.19) | 10.102 | 10.054 | (.10) | 337 |
| ln Urban | 3.481 | 1.063 | (.44) | 4.237 | 1.527 | (.45) | 3.191 | 2.774 | (.07) | 337 |
| Unemployment | .034 | .027 | (.19) | .038 | .026 | (.3) | .028 | .034 | (-.14) | 337 |
| Night lights | .77 | .193 | (.23) | 1.006 | .249 | (.27) | .774 | .577 | (.06) | 337 |
| Poverty Index | .988 | 1.037 | (-.16) | .968 | 1.032 | (-.18) | 1.03 | .990 | (.11) | 337 |
| ln Conflicts | .411 | .217 | (.20) | .454 | .272 | (.17) | .289 | .387 | (-.10) | 337 |
| Bank branches | .866 | .393 | (.25) | 1.054 | .444 | (.28) | .583 | .802 | (.11) | 337 |
Balance table: Municipalities by quartiles of lightning and incidental coverage
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | |
|---|---|---|---|---|---|---|---|---|---|---|
| 25th Percentile | 50th Percentile | 75th Percentile | ||||||||
| PANEL A: Lightning Strikes (Zlight) | ||||||||||
| Below | Above | Dif. | Below | Above | Dif. | Below | Above | Dif. | N. mun. | |
| 2G coverage | .93 | .928 | (.01) | .937 | .92 | (.08) | .924 | .942 | (-.09) | 337 |
| Mean altitude | 1.499 | 1.839 | (-1.06) | 1.582 | 1.922 | (-.91) | 1.673 | 1.988 | (-.69) | 336 |
| Wood | .04 | 3.453 | (-.26) | .024 | 5.175 | (-.32) | .683 | 8.349 | (-.37) | 336 |
| Water | 3.008 | 1.042 | (.26) | 1.629 | 1.437 | (.03) | 1.866 | .535 | (.22) | 336 |
| Sand | 0 | .001 | (-.04) | .001 | 0 | (.08) | .001 | 0 | (.07) | 336 |
| Main Power line | .226 | .246 | (-.03) | .232 | .25 | (-.03) | .222 | .298 | (-.12) | 336 |
| Hospital | .143 | .115 | (.06) | .131 | .113 | (.04) | .119 | .131 | (-.03) | 336 |
| ln Population | 10.175 | 10.03 | (.31) | 10.117 | 10.016 | (.22) | 10.057 | 10.095 | (-.09) | 337 |
| ln Urban | 3.583 | 2.644 | (.15) | 3.622 | 2.138 | (.24) | 2.736 | 3.298 | (-.09) | 337 |
| Unemployment | .036 | .031 | (.13) | .037 | .028 | (.21) | .033 | .031 | (.06) | 337 |
| Night lights | .404 | .700 | (-.12) | .763 | .49 | (.09) | .674 | .484 | (.07) | 337 |
| Poverty Index | 1.049 | .984 | (.16) | .996 | 1.004 | (-.02) | .997 | 1.008 | (-.03) | 337 |
| ln Conflicts | .145 | .435 | (-.31) | .242 | .482 | (-.23) | .318 | .494 | (-.16) | 337 |
| Bank branches | .762 | .743 | (.01) | .863 | .633 | (.10) | .75 | .741 | (0.00) | 337 |
| PANEL B: Incidental Coverage (Zincid) | ||||||||||
| Above | Below | Dif. | Above | Below | Dif. | Above | Below | Dif. | N. mun. | |
| 2G coverage | .956 | .845 | (.49) | .969 | .888 | (.41) | .974 | .913 | (.36) | 337 |
| Mean altitude | 1.734 | 1.806 | (-.16) | 1.693 | 1.810 | (-.27) | 1.669 | 1.779 | (-.23) | 336 |
| Wood | .604 | 8.586 | (-.38) | .447 | 4.726 | (-.27) | .083 | 3.425 | -(.25) | 336 |
| Water | 1.756 | .866 | (.14) | 2.197 | .878 | (.18) | 3.079 | 1.026 | (.24) | 336 |
| Sand | .001 | 0 | (.03) | .001 | 0 | (.07) | 0 | .001 | (-.07) | 336 |
| Main Power line | .274 | .143 | (.23) | .287 | .195 | (.15) | .229 | .245 | (-.03) | 336 |
| Hospital | .139 | .071 | (.16) | .15 | .095 | (.12) | .084 | .134 | (-.11) | 336 |
| ln Population | 10.091 | 9.993 | (.23) | 10.111 | 10.022 | (.19) | 10.102 | 10.054 | (.10) | 337 |
| ln Urban | 3.481 | 1.063 | (.44) | 4.237 | 1.527 | (.45) | 3.191 | 2.774 | (.07) | 337 |
| Unemployment | .034 | .027 | (.19) | .038 | .026 | (.3) | .028 | .034 | (-.14) | 337 |
| Night lights | .77 | .193 | (.23) | 1.006 | .249 | (.27) | .774 | .577 | (.06) | 337 |
| Poverty Index | .988 | 1.037 | (-.16) | .968 | 1.032 | (-.18) | 1.03 | .990 | (.11) | 337 |
| ln Conflicts | .411 | .217 | (.20) | .454 | .272 | (.17) | .289 | .387 | (-.10) | 337 |
| Bank branches | .866 | .393 | (.25) | 1.054 | .444 | (.28) | .583 | .802 | (.11) | 337 |
Sequential g-estimation
Sequential g-estimation
| (1) | (2) | ||
|---|---|---|---|
| Probability of | |||
| Bank Loan | Bank Loan | Total Effect | |
| Std 3G coverage | 0.008*** (0.003) |
0.020* (0.011) |
0.034*** (0.012) |
| Estimation | OLS | IV | IV |
| Borrower FE | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes |
| SW F-statistic | 16.22 | 16.22 | |
| Obs. | 909165 | 909165 | 909165 |
| Adj. R sq. | 0.001 | ||
| Mean Dep. Var. | 0.136 | 0.136 | 0.136 |
| S.D. Dep. Var. | 0.343 | 0.343 | 0.343 |
Sequential g-estimation
| (1) | (2) | ||
|---|---|---|---|
| Probability of | |||
| Bank Loan | Bank Loan | Total Effect | |
| Std 3G coverage | 0.008*** (0.003) |
0.020* (0.011) |
0.034*** (0.012) |
| Estimation | OLS | IV | IV |
| Borrower FE | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes |
| SW F-statistic | 16.22 | 16.22 | |
| Obs. | 909165 | 909165 | 909165 |
| Adj. R sq. | 0.001 | ||
| Mean Dep. Var. | 0.136 | 0.136 | 0.136 |
| S.D. Dep. Var. | 0.343 | 0.343 | 0.343 |
Robustness
Mobile internet and aggregate bank characteristics
Mobile internet and aggregate bank characteristics
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Number of Loans | Number of Switchers | Volume of Loans | Lending Rate | |
| Std 3G coverage | 0.130** (0.065) |
0.118*** (0.034) |
0.285*** (0.091) |
0.279 (0.287) |
| Estimation | IV | IV | IV | IV |
| Bank-municipality FE | Yes | Yes | Yes | Yes |
| Bank-year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 15.97 | 15.97 | 15.62 | 15.70 |
| Obs. | 12695 | 12695 | 12459 | 12467 |
| Mean Dep. Var. | 1.307 | 0.171 | 8.650 | 18.75 |
| S.D. Dep. Var. | 1.301 | 0.420 | 1.696 | 5.375 |
Mobile internet and aggregate bank characteristics
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Number of Loans | Number of Switchers | Volume of Loans | Lending Rate | |
| Std 3G coverage | 0.130** (0.065) |
0.118*** (0.034) |
0.285*** (0.091) |
0.279 (0.287) |
| Estimation | IV | IV | IV | IV |
| Bank-municipality FE | Yes | Yes | Yes | Yes |
| Bank-year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 15.97 | 15.97 | 15.62 | 15.70 |
| Obs. | 12695 | 12695 | 12459 | 12467 |
| Mean Dep. Var. | 1.307 | 0.171 | 8.650 | 18.75 |
| S.D. Dep. Var. | 1.301 | 0.420 | 1.696 | 5.375 |
Mobile internet and access to credit—All borrowers
Mobile internet and access to credit—All borrowers
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.033*** (0.012) |
-0.035*** (0.013) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 15.76 | 15.76 |
| Obs. | 1.268e+06 | 1.268e+06 |
| Mean Dep. Var. | 0.157 | 0.190 |
| S.D. Dep. Var. | 0.363 | 0.392 |
Mobile internet and access to credit—All borrowers
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.033*** (0.012) |
-0.035*** (0.013) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 15.76 | 15.76 |
| Obs. | 1.268e+06 | 1.268e+06 |
| Mean Dep. Var. | 0.157 | 0.190 |
| S.D. Dep. Var. | 0.363 | 0.392 |
Mobile internet and access to credit—First relationship
Mobile internet and access to credit—First relationship
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.048*** (0.015) |
-0.025* (0.014) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 16.22 | 16.22 |
| Obs. | 909165 | 909165 |
| Mean Dep. Var. | 0.163 | 0.189 |
| S.D. Dep. Var. | 0.369 | 0.392 |
Mobile internet and access to credit—First relationship
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.048*** (0.015) |
-0.025* (0.014) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 16.22 | 16.22 |
| Obs. | 909165 | 909165 |
| Mean Dep. Var. | 0.163 | 0.189 |
| S.D. Dep. Var. | 0.369 | 0.392 |
Mobile internet and access to credit—Two-way fixed effects
Mobile internet and access to credit—Two-way fixed effects
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Dummy 3G | 0.094*** (0.036) |
-0.110*** (0.035) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 19.17 | 19.17 |
| Obs. | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 |
Mobile internet and access to credit—Two-way fixed effects
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Dummy 3G | 0.094*** (0.036) |
-0.110*** (0.035) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 19.17 | 19.17 |
| Obs. | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 |
Mobile internet and access to credit—Weighted 3G coverage
Mobile internet and access to credit—Weighted 3G coverage
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of | ||||
| Bank Loan | MFI Loan | Bank Loan | MFI Loan | |
| Std 3G coverage (weighted) | 0.051** (0.022) |
-0.047** (0.020) |
0.049*** (0.017) |
-0.048*** (0.017) |
| Estimation | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 9.669 | 9.669 | 11.63 | 11.63 |
| Obs. | 905574 | 905574 | 905574 | 905574 |
| Mean Dep. Var. | 0.137 | 0.163 | 0.137 | 0.163 |
| S.D. Dep. Var. | 0.343 | 0.370 | 0.343 | 0.370 |
Mobile internet and access to credit—Weighted 3G coverage
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of | ||||
| Bank Loan | MFI Loan | Bank Loan | MFI Loan | |
| Std 3G coverage (weighted) | 0.051** (0.022) |
-0.047** (0.020) |
0.049*** (0.017) |
-0.048*** (0.017) |
| Estimation | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 9.669 | 9.669 | 11.63 | 11.63 |
| Obs. | 905574 | 905574 | 905574 | 905574 |
| Mean Dep. Var. | 0.137 | 0.163 | 0.137 | 0.163 |
| S.D. Dep. Var. | 0.343 | 0.370 | 0.343 | 0.370 |
Mobile internet and access to credit–Alternative versions of the instruments
Mobile internet and access to credit–Alternative versions of the instruments
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of | ||||
| Bank Loan | MFI Loan | Bank Loan | MFI Loan | |
| Std 3G coverage | 0.056*** (0.017) |
-0.086*** (0.022) |
0.047*** (0.015) |
-0.070*** (0.018) |
| Estimation | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 11.55 | 11.55 | 13.06 | 13.06 |
| Obs. | 909165 | 909165 | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 | 0.343 | 0.370 |
Mobile internet and access to credit–Alternative versions of the instruments
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of | ||||
| Bank Loan | MFI Loan | Bank Loan | MFI Loan | |
| Std 3G coverage | 0.056*** (0.017) |
-0.086*** (0.022) |
0.047*** (0.015) |
-0.070*** (0.018) |
| Estimation | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 11.55 | 11.55 | 13.06 | 13.06 |
| Obs. | 909165 | 909165 | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 | 0.343 | 0.370 |
Mobile internet and access to credit—Single instruments
Mobile internet and access to credit—Single instruments
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of | ||||
| Bank Loan | MFI Loan | Bank Loan | MFI Loan | |
| Std 3G coverage | 0.036** (0.015) |
-0.035** (0.015) |
0.018 (0.017) |
-0.037** (0.018) |
| Estimation | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 18.49 | 18.49 | 11.81 | 11.81 |
| Obs. | 909165 | 909165 | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 | 0.343 | 0.370 |
Mobile internet and access to credit—Single instruments
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of | ||||
| Bank Loan | MFI Loan | Bank Loan | MFI Loan | |
| Std 3G coverage | 0.036** (0.015) |
-0.035** (0.015) |
0.018 (0.017) |
-0.037** (0.018) |
| Estimation | IV | IV | IV | IV |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| SW F-statistic | 18.49 | 18.49 | 11.81 | 11.81 |
| Obs. | 909165 | 909165 | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 | 0.343 | 0.370 |
Mobile internet and access to credit—Geographical instruments
Mobile internet and access to credit—Geographical instruments
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.048*** (0.009) |
-0.049*** (0.009) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 861.54 | 861.54 |
| Obs. | 906990 | 906990 |
| Mean Dep. Var. | 0.136 | 0.163 |
| S.D. Dep. Var. | 0.343 | 0.369 |
Mobile internet and access to credit—Geographical instruments
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.048*** (0.009) |
-0.049*** (0.009) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 861.54 | 861.54 |
| Obs. | 906990 | 906990 |
| Mean Dep. Var. | 0.136 | 0.163 |
| S.D. Dep. Var. | 0.343 | 0.369 |
Mobile internet and access to credit—Reduced form
Mobile internet and access to credit—Reduced form
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of | ||||
| Bank Loan | MFI Loan | Bank Loan | MFI Loan | |
| Std Lightning | -0.013** (0.006) |
0.012** (0.006) |
||
| Std Incidental | 0.003 (0.004) |
-0.008* (0.004) |
||
| Altitude | -0.021*** (0.006) |
0.024*** (0.006) |
||
| Wood | 0.001 (0.012) |
0.003 (0.009) |
||
| Water | -0.008** (0.003) |
0.008* (0.004) |
||
| Sand | 0.014*** (0.001) |
-0.008*** (0.001) |
||
| Estimation | OLS | OLS | OLS | OLS |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| Obs. | 909165 | 909165 | 906920 | 906920 |
| Adj. R sq. | 0.372 | 0.369 | 0.374 | 0.370 |
| Mean Dep. Var. | 0.136 | 0.163 | 0.136 | 0.163 |
| S.D. Dep. Var. | 0.343 | 0.369 | 0.343 | 0.369 |
Mobile internet and access to credit—Reduced form
| (1) | (2) | (3) | (4) | |
|---|---|---|---|---|
| Probability of | ||||
| Bank Loan | MFI Loan | Bank Loan | MFI Loan | |
| Std Lightning | -0.013** (0.006) |
0.012** (0.006) |
||
| Std Incidental | 0.003 (0.004) |
-0.008* (0.004) |
||
| Altitude | -0.021*** (0.006) |
0.024*** (0.006) |
||
| Wood | 0.001 (0.012) |
0.003 (0.009) |
||
| Water | -0.008** (0.003) |
0.008* (0.004) |
||
| Sand | 0.014*** (0.001) |
-0.008*** (0.001) |
||
| Estimation | OLS | OLS | OLS | OLS |
| Borrower FE | Yes | Yes | Yes | Yes |
| Year FE | Yes | Yes | Yes | Yes |
| Obs. | 909165 | 909165 | 906920 | 906920 |
| Adj. R sq. | 0.372 | 0.369 | 0.374 | 0.370 |
| Mean Dep. Var. | 0.136 | 0.163 | 0.136 | 0.163 |
| S.D. Dep. Var. | 0.343 | 0.369 | 0.343 | 0.369 |
Mobile internet and access to credit—Additional controls
Mobile internet and access to credit—Additional controls
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.037*** (0.013) |
-0.028* (0.014) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 17.34 | 17.34 |
| Obs. | 931774 | 931774 |
| Mean Dep. Var. | 0.155 | 0.183 |
| S.D. Dep. Var. | 0.361 | 0.386 |
Mobile internet and access to credit—Additional controls
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.037*** (0.013) |
-0.028* (0.014) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 17.34 | 17.34 |
| Obs. | 931774 | 931774 |
| Mean Dep. Var. | 0.155 | 0.183 |
| S.D. Dep. Var. | 0.361 | 0.386 |
Mobile internet and access to credit—Controlling for the U-SACCO program
Mobile internet and access to credit—Controlling for the U-SACCO program
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.035*** (0.012) |
-0.037*** (0.011) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 16.26 | 16.26 |
| Obs. | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 |
Mobile internet and access to credit—Controlling for the U-SACCO program
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.035*** (0.012) |
-0.037*** (0.011) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 16.26 | 16.26 |
| Obs. | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 |
Mobile internet and access to credit—Controlling for broadband internet
Mobile internet and access to credit—Controlling for broadband internet
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.033*** (0.012) |
-0.035*** (0.011) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 16.31 | 16.31 |
| Obs. | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 |
Mobile internet and access to credit—Controlling for broadband internet
| (1) | (2) | |
|---|---|---|
| Probability of | ||
| Bank Loan | MFI Loan | |
| Std 3G coverage | 0.033*** (0.012) |
-0.035*** (0.011) |
| Estimation | IV | IV |
| Borrower FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 16.31 | 16.31 |
| Obs. | 909165 | 909165 |
| Mean Dep. Var. | 0.136 | 0.164 |
| S.D. Dep. Var. | 0.343 | 0.370 |
Broadband internet and land transactions
Broadband internet and land transactions
| (1) | (2) | |
|---|---|---|
| 50% Benchmark | Number of Transactions | |
| Std 3G coverage | 0.162** (0.082) |
0.013*** (0.037) |
| Fiber | 0.002 (0.063) |
-0.024 (0.028) |
| Estimation | IV | IV |
| Municipality FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 19.43 | 19.43 |
| SW F-statistic | 50.83 | 50.83 |
| Obs. | 2022 | 2022 |
| Mean Dep. Var. | 0.121 | 0.229 |
| S.D. Dep. Var. | 0.327 | 0.316 |
Broadband internet and land transactions
| (1) | (2) | |
|---|---|---|
| 50% Benchmark | Number of Transactions | |
| Std 3G coverage | 0.162** (0.082) |
0.013*** (0.037) |
| Fiber | 0.002 (0.063) |
-0.024 (0.028) |
| Estimation | IV | IV |
| Municipality FE | Yes | Yes |
| Year FE | Yes | Yes |
| SW F-statistic | 19.43 | 19.43 |
| SW F-statistic | 50.83 | 50.83 |
| Obs. | 2022 | 2022 |
| Mean Dep. Var. | 0.121 | 0.229 |
| S.D. Dep. Var. | 0.327 | 0.316 |
A municipality refers to the third level administrative subdivision in Rwanda, corresponding to the so-called sector. There are 5 provinces in Rwanda that are subdivided into 30 districts. Each district is in turn divided into sectors. Overall, there are 416 sectors.
The “direct effect” of mobile Internet is a residual effect and catches many potential channels through which the Internet may make individuals seek bank credit: better job opportunities (Hjort and Poulsen, 2019; Caldarola et al., 2023), higher income (Bahia et al., 2023; Calderone et al., 2018), and lower information frictions (Gupta et al., 2023).
It is worth noting that the total number of municipalities in Rwanda is 416, and the missing municipalities in our sample are a result of the inability to differentiate between municipalities with the same name but in different districts within the credit bureau data. Additionally, we exclude observations where borrowers have missing age information to ensure the accuracy of the analysis and avoid including borrowers who may be under the age of 16 in our balanced panel analysis.
Collins Bartholomew is a digital mapping provider.
The GSM network is the dominant standard in Africa and covers around 96% of the market share. These data come from submissions made directly by mobile operators for the purposes of constructing roaming coverage maps for end users.
Data on incidental coverage have been kindly provided by Daniel Bjorkegren.
When examining the effect on bank loan characteristics, the methodology is similar but we use an unbalanced panel dataset based on outstanding relationships.
The choice of the LPM allows to interpret the estimates consistently with the Local Average Treatment Effect (LATE) interpretation (Imbens and Angrist, 1994) and removes the incidental parameters problem, that would be particularly relevant for maximum likelihood estimators in our setting.
The analysis of bank loan characteristics, which focuses on the intensive margin of the credit relationship, uses a specification at the borrower-bank-year level. This allows to include bank-time fixed effects, which further controls for time-varying demand side factors.
The 3G technology was officially introduced in 2010. However, during the first year, less than 8% of the country was connected.
In the robustness section we also adopt for an alternative version of the two instruments, in which
More precisely, the panel is “almost” balanced, since we remove observations for which borrowers are younger than 16 years old.
The p-values of the Hansen J-statistic associated with the specifications in columns 5 and 6 are 0.39 and 0.94, respectively. This indicates that the overidentifying restrictions on the instruments are not rejected, providing further support for the instrument validity.
A municipality is defined to be rural if the share of urban population is equal to zero. A municipality is poor if it is above the median of the poverty index.
The panel dataset is unbalanced and based on outstanding relationships.
We consider the credit characteristics of the borrower, without conditioning on having a loan with a formal bank.
Figure A2 in the appendix plots the distribution of interest rates for switchers, distinguishing between pre and post-switching. The chart shows that borrowers in the upper tail of the distribution particularly benefit from becoming bank customers.
In March 2011, the RNRA launched its Twitter account @Lands Rwanda and created a Facebook page (https://www.facebook.com/LandsRwanda/).
Data are sourced from the Doing Business Indicators published by the World Bank.
Around 80% of all transactions were completed by 2012.
Results are confirmed using a simple dummy variable indicating whether at least one land transaction has been implemented in municipality m, in year t.
The Sobel-Goodman test is an application of the delta method. The idea is that when a mediator is included in a regression analysis with the independent variable, the effect of the independent variable is reduced and the effect of the mediator remains significant. The Sobel-Goodman test is a specialized t-test that provides a method to determine whether the reduction in the effect of the independent variable, after including the mediator in the model, is a significant reduction and therefore whether the mediation effect is statistically significant.
These results are corroborated by the findings in Table A2 in the appendix. Here, we implement sequential g-estimation following Acharya et al. (2016) and show that two-thirds of the total effect of mobile internet on the probability of accessing a bank loan is not mediated through land titles.
This is relevant when we create the balanced panel dataset.
In our case, the monotonicity condition requires all individuals to respond more to lightning strikes than to incidental coverage, or vice versa
The partial monotonicity condition is that the Imbens and Angrist (1994) monotonicity condition is satisfied for each instrument separately, holding all of the other instruments fixed. In our case, a sufficient condition for partial monotonicity is that all individuals are at least as likely to have 3G coverage if they live in municipalities with a low frequency of lightning strikes or in municipalities with higher incidental coverage. However, unlike the Imbens and Angrist (1994) monotonicity condition, partial monotonicity does not restrict heterogeneity in the relative impacts of different instruments; it allows for some individuals to respond more to lightning than to incidental coverage, and for others to respond more to incidental coverage than to lightning
