University of Groningen Faculty of Economics and Business

University of Groningen

Faculty of Economics and Business

To pray for rain or not, that is the question.

performance: empirical evidence from NGOs in Bolivia

Master’s Thesis

Author:

Nazik Asryan

S2805162

Supervisors:

Dr. Fransesco Cecchi

Prof. Dr. Robert Lensink

To pray for rain or not, that is the question.

performance: empirical evidence from NGOs in Bolivia

by Nazik Asryan

Abstract

Using a unique dataset constructed based on data from 10 non-governmental orga-nizations (NGOs), we investigate the relation between weather shocks, microcredit demand, and loan repayment performance in rural Bolivia. The results of the anal-ysis provide evidence that in rural areas households increase borrowings in times of adverse weather events. The results also suggest that rainfall shocks are associated with larger overdue loans. We also observe some evidence on the relation between loan repayment performance and future loan disbursements, though we are not able to fully disentangle supply-side effects to understand the reaction of microfinance institutions (MFIs) to loan portfolio deterioration.

JEL classification: G21, Q54, E51, O13, O16, C33, C35

Contents

1 Introduction 4

2 Literature Review 5

2.1 Weather Risk and Demand for Microcredit . . . 6

2.2 Weather Risk and Repayment Performance . . . 9

2.3 Hypotheses Development . . . 12 3 Data 14 3.1 Data Collection . . . 14 3.1.1 Microfinance Data . . . 14 3.1.2 Weather Data . . . 18 3.2 Data Description . . . 19 3.2.1 Microfinance Data . . . 19 3.2.2 Weather Data . . . 25 4 Empirical Strategy 28 4.1 Variables . . . 28 4.2 Methodology . . . 30 5 Results 32

6 Discussion and Conclusion 39

List of Figures

1 Topographical (a) and political (b) maps of Bolivia . . . 26

List of Tables

2 Number of FINRURAL member NGOs’ branches in nine departments in Bolivia . . . 16

3 The process of dataset construction based on FINRURAL information . . . 17

4 The process of dataset construction based on NOAA information . . . 18

5 Descriptive statistics of rural loan portfolios based on repayment status . . 19

6 Summary of rural loan portfolio quality in 9 departments in Bolivia . . . . 20

7 Descriptive statistics of rural loan portfolios based on repayment status . . 22

8 Descriptive statistics of rural loan portfolios based on targeted sectors . . . 23

9 Descriptive statistics of loan portfolios based on lending types . . . 24

10 Summary of financial situation of FINRURAL member NGOs . . . 25

11 Summary of precipitation data in 9 departments in Bolivia . . . 27

12 Effect of weather shocks on the current rural loan portfolio . . . 33

13 Effect of weather shocks on the overdue rural loan portfolio . . . 35

14 Effect of weather shocks on the share of overdue rural loan portfolio . . . . 37

1

Introduction

“Without water, there is no life” -The response of a farmer to the news of the worst drought in Bolivia in last 25 years1

According to the 2016 report of the Food and Agriculture Organization of the United Nations, up to 122 million more people worldwide could be living in extreme poverty by 2030 as a result of climate change and its impacts on small-scale farmers incomes (FAO, 2016). Dowla (2009) argues that “In Latin America, global warming will lead to the gradual replacement of tropical forest by savanna in eastern Amazonia. Declines in the productivity of major crops and livestock will adversely affect food security. Disappear-ance of glaciers and changes in rainfall will adversely affect agriculture, energy generation, and the availability of water safe for human consumption”. Thurlow et al. (2009) argue that while most of the people directly affected by climate change and extreme weather events live in rural areas, climate variability may also greatly increase urban poverty due to higher food prices and lower real urban incomes. Authors argue that the national poverty rates may significantly rise in particularly severe drought years.

Microfinance offers financial services to poor individuals or households who would otherwise not qualify for financial services including loans, savings, insurance and health saving accounts. Poverty alleviation is the stated goal of many MFIs. In the context of worrying trend of increased poverty because of climate change, accomplishment of the goal may become challenging. In this situation, not only households, but also MFIs should adapt to the changes. The adaptation may involve climate-proofing the existing products and services, developing insurance products and disaster funds. A thorough research is needed to understand the magnitude and channels through which weather-related shocks affect MFIs and what policy solutions are necessary to adapt to them.

Using a unique dataset constructed based on data from 10 NGOs, we investigate the relation between weather shocks, microcredit demand, and loan repayment performance in rural Bolivia. We intend to analyze 3 main questions. The first research question relates to the impact of weather shocks on the rural credit demand. Although borrow-ings have widely been discussed as a coping tool in case of adverse weather events, to the best of our knowledge, their impact on the aggregate demand of credit has not yet been examined. We further analyze whether rural loan repayment rates are negatively affected by adverse weather events. Finally, we examine the reaction of MFIs in Bolivia to the rural loan portfolio deterioration. The results of the analysis provide evidence that in rural areas, households increase borrowings in times of negative rainfall shocks. The results also suggest that rainfall shocks are associated with larger overdue loans. We also observe some evidence on the relation between loan repayment performance and future loan disbursements, though we are not able to fully disentangle supply-side effects to un-derstand the reaction of MFIs to loan portfolio deterioration.

The remainder of this study is organized as follows: Section 2 presents an overview of the literature on the relation between weather shocks, credit demand, and repayment performance. It further discusses the hypotheses tested in the study. Section 3 discusses the data construction process and presents some descriptive statistics. Section 4 elabo-rates on the empirical strategy. The results are presented in Section 5. Conclusions of the study, as well as discussions regarding the limitations, are presented in Section 6.

2

Literature Review

weather-related risks and repayment performance of loans. The section is concluded by the discussion of several hypotheses that are further tested empirically.

2.1

Weather Risk and Demand for Microcredit

Weather shocks are one of the important exogenous factors of riskiness that affect the decision-making process and income variability in rural areas. Several studies have discussed and empirically documented the role of weather shocks and absence of means to cope with it as important factors influencing vulnerability and poverty. Barnett and Mahul (2007) note that poor rural households are particularly sensitive to the conse-quences of weather-related shocks. Most of them have income sources that are tied to the success of agricultural production or are otherwise highly susceptible to extreme weather events such as droughts and floods that can directly destroy productive assets and can indirectly affect poverty of households by enhancing risk-aversion. Mehar et al. (2016) note that agricultural productivity is known to be sensitive to climate change induced effects and it has an impact on the livelihood of families linked with farming.

The availability of rain insurance would also have a positive impact by both raising overall profit levels in high-risk areas and decreasing earnings inequality. According to Mahanta and Das (2017), weather shocks make people risk averse which in its turn generates less income via altering peoples behavior and thus increases their probability of becoming poor. Thus, shocks become a source of vulnerability to poverty.

Poor rural households use different strategies to cope with negative impacts of weather-related shocks. Mahanta and Das (2017) suggest three types of coping mech-anisms: assistance-based measure such as borrowing from different sources, asset-based measures like sale of asset and savings and behavior-based response including participa-tion in government schemes, re-allocaparticipa-tion of labor, devising strategies in an occupaparticipa-tional and livelihood domain. Morduch (1994) argue that second-based arrangements such as asset-base and behavior-base measures as well as borrowing from neighbors and family provide some insurance, they, however, cannot guarantee perfect consumption smoothing. Several studies have shown that such second-best arrangements are mostly imperfect in smoothing weather-related income shocks and many households are severely exposed to them (Dercon and Krishman, 2000; Kazianga and Udry, 2006; Duflo and Udry, 2004; Maccini and Yang, 2009). Udry (1994) and Udry (1995) argue that when insurance mar-kets are incomplete, saving and credit transactions assume a special role by allowing households to smooth their consumption streams in face of random income fluctuations. The author first shows that credit contracts play a direct role in pooling idiosyncratic risk among rural households in Northern Nigeria. Based on a survey data from 200 rural households in Nigeria, he then presents evidence that households reduce their savings by considerable amounts when they receive adverse shocks and they are able to forecast near-future adverse shocks and increase their current savings.

Demont (2013) argues that in presence of credit constraints, microfinance can al-low households to borrow for consumption smoothing in presence of shocks, to avoid asset liquidation and can help households to invest in profitable enterprises. So it is possible that microfinance can shift the means by which households finance consumption smoothing to more efficient methods. The author analyzes household panel data from Indian Self-Help-Groups and weather data to measure the impact of weather shocks on household welfare and the role of microfinance groups. The author documents an ex-tremal vulnerability of farmers to rainfall shocks. The members of Self-Help-Groups had constant access to credit and were able to absorb the adverse effects of shocks. Lensink et al. (2017) examine the impact of membership in an MFI on the vulnerability to poverty in Mexico. Authors argue that membership in MFIs improves income and decreases income variability reducing the probability of becoming poor. They also show that households that use microfinance services are able to better smooth consumption in the events of adverse shocks.

Summarizing, the literature suggests that weather-related shocks are important fac-tors affecting income, consumption, and welfare of rural households. Credit-constrained households use various informal tools to cope with negative effects of weather shocks, which, however, do not provide perfect consumption smoothing. In this situation, mi-crofinance may provide an effective solution to cope with adverse events and reduce the vulnerability of poor households.

2.2

Weather Risk and Repayment Performance

because of the inability of borrowers to repay their loans. In developing countries, the negative effect of natural disasters on financial institutions is stronger as fewer households are insured against weather-related risks and portfolios have greater economic sector and geographic concentration. Authors argue that natural disasters affect the financial per-formance of institutions in various ways by not only destroying the value of current loans but also constraining provision of future loans. Bad repayment performance of borrowers may adversely affect the revenues of financial institutions and may result in increased level of non-performing loans. It may result in increased administrative costs because of loan restructuring and collateral exercising.

Moser and Gonzalez (2015) discuss the effect of climate change on MFIs. Authors argue that less-developed countries are mostly exposed to the extreme weather shocks caused by climate change. Microfinance clients globally are the most vulnerable to cli-mate variability (Stern, 2007; Ahmed et al., 2009). According to Dowla (2009), increased incidence of extreme weather events such as droughts, flooding, and storms are very likely to endanger assets such as homes, crop yields, micro-businesses, and livestock of many microfinance clients worldwide, and thus, undermine their ability to repay loans. Extreme weather events may also have a direct impact on physical infrastructure such as equip-ment, offices, records and information systems of MFIs.

Pelka et al. (2015) empirically investigate the effect of weather-related shocks on fi-nancial outcomes in microfinance. They use a detailed lending data provided by an MFI in Madagascar and investigate the effect of weather variation on the repayment performance of loans granted to small-scale farmers. They show that excessive amount of precipitation in the harvesting period of rice increases the credit risk of loans granted to small-scale farmers in Madagascar.

in Peru. El Ni˜no creates abnormal levels of precipitation and flooding in northern Peru resulting in damaged infrastructure, destroyed assets, and crop. Several studies docu-ment the negative effect of 1998 events on loan repaydocu-ment and interest rates. According to Trivelli (2006), the El Ni˜no created long-lasting loan repayment problems in Peru. Collier et al. (2011) analyze the loan portfolio performance of a rural lender in Peru after the flooding of 1998 and conclude that the event significantly increased the proportion of restructured loans which are an important measure of loan portfolio quality as they indi-cate the inability of borrowers to repay. The negative impact was especially pronounced because of correlated risk exposure of many small-holders.

The dramatic events of 1998 resulted in credit rationing in agriculture in exposed regions in Peru. The limited credit combined with physical damage caused by flooding created additional pressure on the welfare of small-holders. Collier et al. (2013) develop a model of the behavior of the lender calibrated using data from a Peruvian financial institution that specializes in microfinance and is vulnerable to the risk of severe El Ni˜no. Authors demonstrate that loan losses from the event destroy the capital of the lender. After the disaster, the lender reduces loan disbursements, bringing them in line with a smaller capital base but also limiting access to credit for borrowing households and firms. Because of the business disruptions created by disaster losses, the risk of these events leads the lender to maintain higher capital reserves reducing the supply of credit in non-disaster conditions and limiting profits and growth of the financial institution.

2.3

Hypotheses Development

A substantial body of literature discusses borrowing as one of the common coping mechanisms in the event of adverse weather events among poor rural households. Several empirical studies provide evidence that availability of microfinance borrowings positively affects the ability of exposed households to cope with disasters (both weather-related and other shocks ) and to smooth consumption without liquidation of assets (Udry, 1994; Pitt and Khandker, 2002; Del Ninno et al., 2003; Kaboski and Townsend, 2005; Khandker, 2007; Gertler et al., 2009; Islam and Maitra, 2012; Demont, 2013). In the same time, however, there have little been said about the effect of adverse shocks on the aggregate demand of microcredit. We suggest testing whether weather-related shocks create credit demand based on the above-mentioned argument that poor rural households use bor-rowing as a consumption-smoothing tool in the events of adverse weather shocks. The Demand hypothesis predicts a positive association between weather shocks and credit demand.

H1-Demand hypothesis: Weather shocks increase credit demand in rural areas. Rural households use borrowings not only in case of adverse events for consump-tion smoothing. They may also require funding for investments in agricultural equip-ment, chemicals, land, and livestock as well as other enterprises and non-business related projects such as house renovations, health care, and education fees. Literature provides evidence on the impact of weather shocks on risk attitudes of rural households. Thus, in the events of an adverse shock, rural households may postpone large investments in businesses and secondary expenditures and cut their demand for credit. The alternative hypothesis, in this case, predicts a negative association between weather shocks and credit demand.

of 1998 El Ni˜no in Peru and find a negative impact on loan portfolio performance. El Ni˜no is, however, one of the various weather-related shocks that can occur. Authors mention that the effects of droughts may be different from the effect of flooding as droughts are less likely to cause infrastructure damage. Given the scarce empirical literature on the topic, we intend to investigate the relation between loan repayment performance and rainfall shocks of different magnitudes and direction. Based on the above-mentioned arguments, we suggest testing whether excessive rainfalls create mobility problems and make it dif-ficult for farmers to travel to MFI branches for loan repayment and whether negative rainfall shocks create repayment problems by damaging crops and decreasing income of farmers.

H2-Mobility hypothesis: Excessive rainfalls are associated with poorer repayment per-formance.

H3-Ability hypothesis: Negative rainfall shocks are associated with poorer repayment performance.

six-month extensions of repayment schedules and flexible savings withdrawal terms. In the context of scarce mixed evidence, we suggest testing whether there are supply-side effects in the event of weather shocks among NGO lenders in Bolivia.

H4-Supply hypothesis: Poor loan repayment performance resulting from weather shocks negatively affects the amount of new loan disbursements.

Summarizing, the following study intends to analyze 3 main questions. Do weather shocks increase rural credit demand? Are rural loan repayment rates negatively affected by adverse weather shocks? If yes, how do the MFIs in Bolivia react to the rural loan portfolio deterioration?

3

Data

In the following section, we present the sources and characteristics of the data used in our study. We use a unique microfinance dataset and present the detailed process of its construction. Further, the weather data collection process is presented. It is followed by the description of variables used in the research and descriptive statistics of the sample.

3.1

Data Collection

3.1.1 Microfinance Data

promote actions that favor the development of microfinance and finance for development and to support the institutional strengthening of its partners. FINRURAL specializes in microfinance and groups 10 NGOs engaged in microfinance. Table 1 provides general information about the 10 member NGOs.

Table 1: General information on FINRURAL memeber NGOs

This table summarizes general information on FINRURAL member NGOs. The year of foundation, the number of clients and branches are presented. Last two columns present the number of urban and rural branches separately.

Name Founded Clients Branches Urban Rural

CIDRE 1981 17,192 29 14 15 CRECER 1999 177,791 83 42 41 DIACONIA-FRIF 1991 65,904 59 37 22 EMPRENDER 1997 10,220 12 10 2 FONDECO 1995 6,379 14 5 9 FURBODE 1997 41,126 20 8 12 IDERPO 1988 12,803 24 12 12 IMPRO 1995 2,400 3 2 1 PRO MUJER 1991 122,714 55 17 8 SEMBRAR SARTAWI 1989 32,236 44 23 21 Source: FINRURAL

Note: Number of clients is as of 02.2017 except for EMPRENDER (08.2016).

This information is presented on department (“departamento” in Spanish) level. There are 9 departments in Bolivia so reports provide a dataset of 3,060 observations. The 9 departments in Bolivia are divided into 112 provinces. The most densely populated departments are Cochabamba and La Paz. The distribution of branches of FINRURAL member NGOs in the country is summarized in Table 2. The greatest number of branches operates in departments with the largest population (La Paz, Santa Cruz, Cochabamba). In the same time departments with the highest level of poverty are Potosi, Chuquisaca, and Pando. In these 3 departments there only 48 operating branches.

Table 2: Number of FINRURAL member NGOs’ branches in nine departments in Bolivia

This table presents the number of FINRURAL member NGOs’ branches in nine departments in Bolivia. Number of urban and rural branches is presented separately in columns 2 and 3. The population and area of each department are presented in column 5 6 respectively

Department Urban Rural Total Population Area (sq.km)

Beni 5 11 16 455,928 213,564 Chuquisaca 13 8 21 616,073 51,524 Cochabamba 18 26 44 1,915,621 55,631 La Paz 79 35 114 22,842,031 133,985 Orudo 14 6 20 525,675 53,588 Pando 4 2 6 133,966 63,827 Potosi 10 11 21 873,901 118,218 Santa Cruz 45 33 78 3,078,459 370,621 Tarija 12 11 23 543,405 37,623

Source: FINRURAL, www.ine.gob.bo (Instituto Nacional Estadistica). Note: Population of departments in Bolivia as of year 2016.

Table 3: The process of dataset construction based on FINRURAL information

Steps Process description

Step 1 Received 34 .xlsx format files from FINRURAL including 32 different sheets of information each.

Information on department level

Step 2 Changed the names of files to a standard format: “Name Year Month”. Step 3 Extracted the relevant cell-ranges from each file and saved in separate

.dta format files.

Step 4 Merged generated .dta files for each MFI. Step 5 Appended the data for all 10 MFIs. Number of observations: 3,088

Step 6 Standardized the format of Date and ID for all MFIs.

Step 7 Manually corrected the mistake in one of the received .xls files for Sartawi and removed duplicate values.

Step 8 Translated names of 270 department-level variables from Spanish. Step 9 Renamed variables and created lables for further reference.

Number of observations: 10 MFIs × 34 Months × 9 Departments = 3,060 Information on country level

Step 10 Standardized accounting codes of items (changed “.” to “ ”)

in the sheets containing country-level information from balance sheets and income statements.

Step 11 Extracted relevant cell-ranges from each sheet of 34 .xls files. Step 12 Merged information for each MFI.

Step 13 Appended data for all MFIs.

Step 14 Standardized the format of Date and ID for each MFI.

Step 15 Dropped all empty variables leaving 344 variables in the dataset. Number of observations: 10 MFIs × 34 Months = 340

Step 16 Appended microfinance data on country and department level.

3.1.2 Weather Data

We have received the data on precipitation from The Physical Sciences Division (PSD) of Earth System Research Laboratory (ESRL)2. ESRL is a laboratory in National Oceanic and Atmospheric Administration’s (NOAA) Office of Oceanic and Atmospheric Research (OAR). The NOAA ESRL Physical Sciences Division (PSD) conducts weather and climate research to observe and understand Earth’s physical environment, and to im-prove weather and climate predictions on global-to-local scales. From the climate datasets, we extracted monthly global gridded precipitation means with a temporal coverage from January 1979 to May 2017 and a spatial coverage of 2.5-degree latitude × 2.5- degree longitude global grid. Table 4 summarizes the weather dataset construction process.

Table 4: The process of dataset construction based on NOAA information

Steps Process description

Step 1 Extracted the .nc format file of standard monthly mean precipitation data for the whole temporal and spatial coverage.

Step 2 Identified 19 grid references (coordinates) in the area of Bolivia and separated the relevant part of the initial dataset.

Step 3 Identified departments in which each coordinate lies.

Step 4 Calculated the average of mean precipitation values for each department using the the values at multiple coordinates in one department (if such existed).

Step 5 Calculated the long-term monthly mean precipitation values.

Step 6 Calculated the monthly deviations from long-term mean precipitation values.

Step 7 Separated the part of the dataset with temporal coverage from October 2011 to July 2016.

Number of observations: 9 Departments × 34 Months = 306.

Source: Data manipulations are done by author using Python programming language and Stata.

2_{CMAP Precipitation data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from}

The average precipitation values are obtained from different satellite estimates and gauge data. The data files are in Network common data form (NetCDF) format. We have extracted the monthly mean precipitation data in the area of Bolivia for the period for which we have the microfinance data from FINRURAL. We have further calculated long-term mean precipitation values for each month using the whole available dataset from January 1979. The obtained values were further used to construct monthly rainfall deviations from long-term mean values. The obtained weather dataset was then appended to the microfinance dataset.

3.2

Data Description

3.2.1 Microfinance Data

The dataset provides information on a wide range of variables. An extensive part of the reports covers details about the loan portfolios of institutions.The total loan portfolios of institutions are first disaggregated based on the repayment status. Table 5 presents the details for rural loan portfolios on the country level. The table for urban loan portfolio is available in Appendix A.

Table 5: Descriptive statistics of rural loan portfolios based on repayment status

This table presents the mean, standard deviation, minimum and maximum values of rural loan portfolios of FINRURAL member NGOs over the period Oct.2013-Jul.2016 disaggregated by the repayment status.

Variable Mean Std.Dev. Min Max

Gross Rural Loan Portfolio 162.00 192.00 1.12 812.00 Current Rural Loan Portfolio 159.00 190.00 0.12 800.00 Restructured Current Rural Loan Portfolio 0.74 0.94 0.00 3.58 Overdue Rural Loan Portfolio 1.59 1.56 0.00 9.56 Restructured Overdue Rural Loan Portfolio 0.09 0.16 0.00 0.81

Source: Author’s calculations based on FINRURAL data.

The size of gross loan portfolios is almost similar for urban and rural branches. For both, the greatest part of the gross loan portfolio is the current portfolio. Around 2% of the gross rural portfolio and 1.8% of the gross rural portfolio is overdue (on average 1.59 million is overdue from the average gross rural portfolio of 162 million Boliviano).

The loan portfolio quality measured by share of overdue and restructured loan port-folios in different departments in Bolivia is summarized in Table 6.

Table 6: Summary of rural loan portfolio quality in 9 departments in Bolivia

This table presents the means and standard deviations of 3 portfolio quality ratios in 9 departments in Bolivia: share of overdue rural loan portfolio in gross rural loan portfolio, share of restructured rural loan portfolio in gross rural loan portfolio an share of restructured overdue rural loan portfolio in gross rural loan portfolio.

Department Share of Overdue Share of Restructured Share of Restructured Rural Loan Rural Loan Overdue Rural

Portfolio (%) Portfolio (%) Loan Portfolio (%)

Beni 4.738 0.689 0.161 (13.931) (0.662) (0.231) Chuquisaca 0.887 0.887 0.496 (1.135) (1.580) (1.804) Cochabamba 2.447 0.365 0.031 (9.187) (0.484) (0.754) La Paz 1.449 0.368 0.075 (2.868) (0.374) (0.102) Oruro 0.961 0.236 0.017 (2.835) (0.341) (0.063) Pando 6.403 0.641 0.080 (19.224) (1.005) (0.181) Potosi 1.303 0.870 0.273 (3.032) (1.227) (0.431) Santa Cruz 2.232 1.734 0.337 (7.305) (2.017) (0.735) Tarija 2.486 1.827 0.194 (7.551) (2.658) (0.438)

The share of loans in the restructured category is calculated by dividing the value of re-structured loans by the value of gross loan portfolio. Rere-structured loans represent losses in outstanding principal and increased costs including the added administrative costs associated with restructuring the loan. When borrowers have difficulties with loan re-payment, financial institutions try to reduce losses and allow borrowers to repay under different terms through restructuring. Restructured loans give an indication about the current losses in assets but also about the opportunity costs related to holding poorly performing loans. Consequently, any changes in the proportion of restructured loans are likely to have longer-term costs for the financial institution. The share of restructured loans is an indicator of the asset quality of the financial institution (Collier et al., 2011). The share of overdue loans is another indicator of asset quality. It is calculated by divid-ing the value of overdue loans by the value of gross loan portfolio. Overdue loans cause monthly revenue losses for financial institutions and increased provisions and reduce the likelihood of repayment. Collier et al. (2011) note that although financial institutions can not directly control whether borrowers are repaying their loans or not, they can choose what proportion of their loans to restructure and thus have some control over the overall level of overdue loans in the portfolio. The largest average share of the overdue rural loan portfolio is documented in Pando, one of the poorest departments in Bolivia (you can find the details for urban loan portfolio in Appendix A).

Table 7: Descriptive statistics of rural loan portfolios based on repayment status

This table summarizes the rural loan portfolios of FINRURAL member NGOs over the period Oct.2013-Jul.2016 disaggregated by the loan sizes. Columns 2 and 3 present the means and standard deviations of rural loan portfolio while columns 4 and 5 present means and standard deviations of number of clients in different loan size ranges.

Loan Portfolio Number of clients

Range Mean Std.Dev. Mean Std.Dev

Greater than USD 200,001 0.17 1.19 0.11 0.76 USD 100,001 - USD 200,000 2.37 6.41 3.98 9.21 USD 50,001 - USD 100,000 7.64 12.6 21.25 34.19 USD 30,001 - USD 50,000 8.76 12.40 46.03 145.72 USD 20,001 - USD 30,000 7.93 9.35 66.82 102.77 USD 15,001 - USD 20,000 7.04 7.33 86.55 116.81 USD 10,001 - USD 15,000 18.50 28.00 348.09 638.27 USD 5,001 - USD 10,000 41.20 49.60 1,230.90 1,786.41 USD 4,001 - USD 5,000 15.60 17.50 650.94 811.56 USD 3,001 - USD 4,000 14.40 14.00 815.30 926.16 USD 2,001 - USD 3,000 22.30 27.70 1,751.72 2,522.29 USD 1,001 - USD 2,000 23.60 30.80 3,319.52 5,244.05 USD 501 - USD 1,000 15.70 24.20 4,437.95 7,956.42 Smaller than USD 500 7.65 11.00 4,803.01 7,798.62

Source: Author’s calculations based on FINRURAL data.

Note: Loan portfolio values and standard deviations are presented in millions Boliviano.

“Services of extraterritorial organizations and bodies” and “Atypical activities” (See Ta-ble 8 for details on rural loan portfolio).

Table 8: Descriptive statistics of rural loan portfolios based on targeted sectors

This table presents the mean, standard deviation, minimum and maximum values of rural loan portfolios of FINRURAL member NGOs over the period Oct.2013-Jul.2016 disaggregated by the targeted sectors. Rows 1 and 2 summarize the loan portfolio of productive and non-productive sectors while rows 3 and 4 summarize the number of operations.

Variable Mean Std.Dev. Min Max

Total Rural Loan Portfolio 109.00 82.50 0.00 297.00 to Productive Sectors

Total Rural Loan Portfolio 83.80 132.00 0.18 529.00 to Non-Productive Sectors

Total Number of Rural Operations 4,758.75 3,930.61 0.00 13,501.00 in Productive Sectors

Total Number of Rural Operations 12,662.97 25,136.40 19.00 91,315.00 in Non-Productive Sectors

Source: Author’s calculations based on FINRURAL data.

Note: Loan portfolio values are presented in millions Boliviano (rows 1 and 2)

The average rural loan portfolio targeted to the productive sectors is larger than the loan portfolio targeted to the non-productive sectors while the average number of operations is significantly smaller. This may be explained by nature of the operations in different sectors. Clients in retail commerce which is a non-productive sector, for example, can use small inventory loans on a weekly or monthly basis, while clients in agriculture may need relatively larger amounts a few times during the year for equipment, crop sup-plies, chemicals or land improvements. This difference between the number of operations is even more pronounced in the urban market (see Appendix A for details on urban loan portfolio).

soli-darity groups. MIXMarket Glossary defines solisoli-darity group lending as a procedure when groups are used for disbursement of funds and collection of repayment on loans to either the group as a whole or to the individual members of that group. Borrowers of such groups often bear joint liability for the repayment of all loans to the group and its mem-bers. Solidarity group lending was first applied by Muhammad Yunus in Grameen Bank. The model with several adjustments was later adopted in Latin America. MIXMarket defines village Banking (as well as Self Help Groups) as methodologies that provide access to credit and savings through group or community managed associations. MFIs lend to the group, which in turn uses this money to lend to its members. This type of loans is made under the collective guarantee of the group. Table 9 provides detailed information on loan portfolios disaggregated by lending types. FINRURAL member NGOs disburse almost all loans through either individual contracts or village banking. Relating to the gender distribution of client base, on average 74% of clients of NGOs are female, from whom 63% live in urban areas. In comparison, only 54% of male clients live in urban areas.

Table 9: Descriptive statistics of loan portfolios based on lending types

This table presents the mean and standard deviation of loan portfolios, number of operations and number of borrowers of FINRURAL member NGOs over the period Oct.2013-Jul.2016 disaggregated by lending methodology.

Disbursement Operations Borrowers Mean Std.Dev. Mean Std.Dev. Mean Std.Dev Solidarity Group 0.03 0.16 7.00 23.74 16.86 140.09

Association 0.00 0.01 0.10 0.61 -

-Individual 17.80 16.60 1,830.78 3,033.50 1,028.44 1,937.66 Village Banking 22.30 37.40 4,020.70 7,361.56 3,424.33 7,184.44

Source: Author’s calculations based on FINRURAL data.

Note: Disbursements are presented in millions Boliviano (columns 1 and 2)

USD3.6m-USD315.5m). In the period discussed, the average value of assets is 436m Bo-liviano. The average Debt to Equity ratio is 5.44. On average, the NGOs are operating with positive profits.

Table 10: Summary of financial situation of FINRURAL member NGOs

This table presents the mean, standard deviation, minimum and maximum values of several indicators from balance sheets and income statements of FINRURAL member NGOs over the period Oct.2013-Jul.2016.

Variable Mean Std.Dev. Min Max

Total Assets 436.00 381.00 26.10 1,540.00 Total Liabilities 347.00 303.00 20.70 1,490.00 Total Equity 88.80 104.00 4.77 377.00 Debt to Equity 5.44 2.420 1.13 28.33 Total Income 58.80 67.50 0.40 378.00 (Total Expenses) (54.90) 62.80 (362.00) (0.39) Gross Revenue 36.80 45.20 0.27 249.00 Gross Operating Profit 37.50 44.20 0.31 245.00 Net Operating Profit 4.28 8.05 -4.86 40.20 Before-tax Profit 3.89 7.32 -4.86 40.40

Net Profit 4.32 8.16 -4.86 43.00

Source: Author’s calculations based on FINRURAL data.

Note: Numbers are presented in millions Boliviano. BOB 1= USD 0.14.

3.2.2 Weather Data

The country profile developed by the Federal Research Division of the Library of Congress (Library of Congress Federal Research Division, 2006) summarizes the climate of Bolivia. Bolivian topography is distinguished by extreme elevation changes. The country is split into three topographical regions: the Andes and arid highlands of the west, the semi-tropical valleys in the middle third of the country, and the tropical lowlands of the east (See Figure 1a 3 _{for the topographical map of Bolivia). This extreme range of}

elevations gives the country a wide range of climates. In the highlands of Bolivia, the rainy season lasts from December until March. The lower, eastern slopes of the Cordillera Oriental, known as the Yungas, compose the semi-tropical region of the country with a rainy season from March to April. By contrast, the Bolivian lowlands in the east, face semi-arid conditions and is the wettest region. The rainy season here lasts from late September until May.

(a) (b)

Figure 1: Topographical (a) and political (b) maps of Bolivia

Since 2000, El Ni˜no events have often occurred with the strongest effects in years from 2014 to 2016.

The weather data obtained from NOAA covers 10,368 coordinates worldwide, of which 19 are in the area of Bolivia. The coordinates are market with red rectangles on Figure 1b. Comparison of Figures 1a and 1b shows that the coordinates for which weather data is available are not evenly distributed among 9 departments in Bolivia. The largest number of coordinates (6) lies in the department of Santa Cruz as it has the largest area. For the departments for which there were more than one coordinate available, we have calculated the average monthly precipitation as the arithmetical mean of monthly mean precipitation values of different coordinates. The summarized statistics in Table 11 show that Oruro, Chuquisaca, and Tarija are the driest departments located in highlands, while departments in Amazon valley have the highest values of rainfall.

Table 11: Summary of precipitation data in 9 departments in Bolivia

This table summarizes the precipitation data in 9 departments in Bolivia. Means, standard deviations, minimum and maximum values of rainfall are presented for the period from October 2011 to July 2016 for 340 observations.

Department Mean Std. Dev. Min Max

Beni 4.642 3.248 0.32 12.255 Chuquisaca 2.670 1.681 0.15 5.88 Cochabamba 4.041 2.869 0.43 11.86 La Paz 3.709 2.848 0.545 14.02 Oruro 0.676 0.676 0.01 2.93 Pando 5.153 3.412 0.175 12.13 Potosi 0.859 0.999 0.02 3.653 Santa Cruz 3.109 2.073 0.119 7.458 Tarija 2.124 2.176 0 8.76

4

Empirical Strategy

We have discussed 4 hypotheses, that we test in this study. First, Demand hypothesis, to check whether weather-related shocks increase rural credit demand. Second, Mobility and Ability hypotheses, to analyze whether adverse weather events negatively affect rural loan repayment performance. Third, Supply hypothesis, to analyze the impact of poor repayment performance on future credit supply. In this section, we discuss the variables and methodology used in the analysis.

4.1

Variables

Given the climate characteristics of Bolivia, we consider rainfall as the most rele-vant weather indicator. There were several incidents of drought in Bolivia during recent years, caused by melting glaciers in Andean mountains and severe El Ni˜no in 2015-2016. According to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC, 2007), the rapid melting of glaciers will decrease the availability of water and small increases in temperature will decrease crop productivity, increasing the risk of hunger. Crop production will be adversely affected by more frequent droughts and floods. The impacts of climate will be felt unevenly across the world, especially affecting Africa, Asia, and Latin America, regions that have the least adaptive capacity enduring the worst consequences. Given that less than 1 percent of agricultural lands are irrigated in Bolivia, the sector of the economy that employs one-third of the countries population is highly dependent on rainfall.

it in terms of standard deviations (see Figure A1 in Appendix A for the distribution of DIFF STD). We further introduce variables indicating rainfall shocks to capture the im-pact of extreme rainfall events. SHOCK POS is a dummy variable indicating positive rainfall shock. It takes value 1 if the difference between actual and long-term average rainfall exceeds 1 standard deviation and 0 otherwise. SHOCK NEG is a dummy variable indicating negative rainfall shock. It takes value 1 if the difference between actual and long-term average rainfall is below -1 standard deviation and 0 otherwise.

The FINRURAL dataset provides detailed information on loan portfolios of member NGOs. Based on this data, we construct several dependent variables. In order to test the Demand hypothesis, we look at the current loan portfolio of NGOs. CURRENT RUR is the monthly per-department current outstanding balance of the rural loan portfolio. There are various factors that can influence the demand of credit, such as age, educational level, marital status and income level of borrowers. Given the nature of our dataset, we cannot control for personal characteristics of borrowers. On the institutional level, the main factor influencing the demand is the interest rate. Taking into account the fact that in the small time period under consideration the interest rates charged by different NGOs have almost no time variation, we do not control for interest rates in regression analysis. In order to test the Mobility and Ability hypotheses, we include the overdue loan portfolio in the analysis. OVERDUE RUR reflects the amount of non-performing loans in the rural loan portfolio. It shows the monthly per-department overdue balance of the rural loan portfolio.

4.2

Methodology

In order to test the effect of rainfall shocks on the current rural loan portfolio and its repayment performance, we perform a set of regressions with the following general form:

Yi,t =α1i+ β1RAINi,t+ γ1L.RAINi,t+ u1i,t (1)

Yi,t =α2i+ β2SHOCK N EGi,t+ γ2L.SHOCK N EGi,t (2)

+ β3SHOCK P OSi,t+ γ3L.SHOCK P OSi,t+ u2i,t,

where Yi,t =Y1i,t, Y2i,t, Y3i,t is a vector of dependent variables with Y1i,t being the current

loan portfolio, Y2i,t being the overdue loan portfolio and Y3i,t being the share of overdue

loans in the gross loan portfolio. β1 = {β11, β12, β13} is a vector of coefficients of monthly

average precipitation rate, β2 = {β21, β22, β23} is a vector of coefficients of the negative

rainfall shocks and β3 = {β31, β32, β33} is a vector of coefficients of the positive rainfall

shocks (second subscripts relate to one of 3 dependent variables). We also include the first and second lags of independent variables as rainfall shocks may have delayed impact. For example, drought or excessive rainfall in a given month may affect the harvest and income of the farmer in one or two months resulting in increased or decreased credit demand. γ1,

γ2 and γ3 are the vectors of coefficients of lagged independent variables.

The Demand hypothesis predicts negative effect of increased rainfall on the current loan portfolio (β11, γ11 < 0) while the coefficient of negative rainfall shocks is expected

to be positive (β21, γ21 > 0) as low levels of rainfall are associated with poor future

be positive (β22, γ22 > 0). The Mobility hypothesis predicts that positive rainfall shocks

have a negative effect on the repayment performance increasing the overdue loan portfolio (β32, γ32 > 0). This is explained by the possibility of damaged infrastructure (e.g. flooded

roads) in case of abnormal excessive rainfalls. In order to check the validity of results, we perform a placebo test and check the Demand, Ability and Mobility hypotheses for urban loan portfolio. As long as we have no reason to believe that rainfall affects the credit demand and repayment performance of urban clients, the coefficients of rainfall variables are expected to have no significant impact on current and overdue urban loan portfolios. The nature of the phenomena under consideration assume a dynamic panel data. The dependent variables in our study, namely the levels of current and overdue loan port-folios, are dynamic, depending on their own past realizations. The independent variables, namely the rainfall shocks, being completely exogenous to the dependent variables, are still not strictly exogenous, in the sense that they can also be correlated with their past. Given the dynamic nature of the panel, as well as the smaller number of time inter-vals compared to cross-sectional units, we apply system generalized method of moments estimation as a robustness check. System GMM estimator was developed by Blundell and Bond (1998). It uses lags of the dependent variable as instruments. We use the “xtabond2” package developed by Roodman (2006) for Stata statistical package.

The Supply hypothesis relates to the relation of loan repayment performance due to weather shocks and new loan disbursements. It intends to analyze the supply-side effects of rainfall shocks. In order to examine the reaction of lenders to the poor repayment per-formance due to weather shocks, we instrument the amount of overdue loans by negative rainfall shocks and perform the following first-stage estimation:

L.OV ERDU E RU Ri,t =λ1i + λ2SHOCK N EGi,t+ 1i,t (3)

an instrument for economic growth. Fichera and Savage (2015) instrument income with rainfall to analyze the impact on health outcomes. The second-stage equation estimates the impact of overdue loans on the amount of new loan disbursements.

N EW LOANi,t =λ3i+ λ4L.OV ERDU E RU Rˆ i,t + 2i,t, (4)

where L.OV ERDU E RU Rˆ i,t is the estimated value of lagged overdue loan portfolio

from the first-stage equation. N EW LOAN is measured as the growth of monthly per-department current loan portfolio. We consider the growth of current loan portfolio a good proxy for the amount of new loan disbursements. Current loans represent the amount of outstanding loans on which the interest is being paid. From the accounting perspective, the change of the amount may also reflect the part of loans which have become overdue between 2 periods and may not reflect the part of loans which were newly disbursed but became overdue until the end of the second period. Nevertheless, given that in absolute terms the amount of overdue loans is small relative to the current portfolio, we consider N EW LOAN an adequate measure of new loan disbursements.

There are several potential concerns with the identification strategy. The first is the exclusion restriction. We consider our identification strategy to be valid as long as we do not have any reason to believe that weather shocks may affect the amount of new loan disbursements via a channel other than the repayment performance of borrowers. Another concern is the possibility of endogeneity, however our measure of weather shocks is based on the average rainfall level which is strictly exogenous and can not be affected by repayment performance of borrowers.

5

Results

NGOs in Bolivia. Table 12 suggests evidence in favor of the Demand hypothesis. Negative significant coefficients of rainfall indicate that higher values of precipitation are associated with smaller current loan portfolio taking into account the seasonal differences and overall yearly portfolio increase.

Table 12: Fixed effects estimation results: Effect of weather shocks on the current rural loan portfolio

This table summarizes the results of fixed effect estimation of rainfall shocks and the current rural loan portfolio. For each variable the coefficient and its standard error are presented. Model 1.1 and 1.2 estimate the impact of the absolute level of rainfall on the current rural loan portfolio. Models 1.3 and 1.4 estimate the impact of negative and positive rainfall shocks.

Model 1.1 Model 1.2 Model 1.3 Model 1.4 RAIN -428,395.0*** -235,541.6** (91,197.6) (93,961.4) L.RAIN -78,823.8** -192,374.8*** (30,600.1) (62,004.8) L2.RAIN -262,877.1*** -145,740.8** (55,324.5) (62,727.9) SHOCK NEG 1,985,909.7*** 824,835.3* (461,881.1) (478,071.2) L.SHOCK NEG 1,794,670.3*** 1,132,946.6** (480,970.6) (449,133.5) L2.SHOCK NEG 2,255,822.7*** 1,294,179.3** (601,202.9) (566,018.5) SHOCK POS -1,293,041.9*** -141,068.9 (356,659.4) (289,991.3) L.SHOCK POS -721,093.4*** -154,961.6 (268,777.0) (262,112.2) L2.SHOCK POS -756,290.8*** -158,292.9 (274,962.5) (275,241.9) Constant 31,755,742.8*** 22,985,501.2*** 29,477,702.0*** 19,897,668.9*** (441,550.4) (1,295,207.1) (245,319.9) (1,710,353.5) Observations 1,744 1,744 1,744 1,744 R-squared 0.1 0.3 0.1 0.3 No. of IDs 58 58 58 58

MFI FEs Yes Yes Yes Yes

Province FEs Yes Yes Yes Yes

Time FEs No Yes No Yes

Negative rainfall shocks are associated with larger current portfolio values while the opposite effect of positive shocks disappears after controlling for monthly and yearly trends. The point estimate of SHOCK NEG shows that in the event of negative rain-fall shocks the per-department current rural loan portfolio of NGOs is on average larger by around 825 thousand Boliviano (around 116 thousand USD). The monthly average per-department current rural loan portfolio of NGOs over the period Oct.2013-Jul.2016 is around 30 million Boliviano, thus negative rainfall shocks on average result in a 2.75% increase in the monthly per-department current rural loan portfolio. As long as rain-fall is an exogenous factor and we have no reason to assume that it can affect lending policy of MFIs, the fixed effect estimation gives the effect of rainfall shocks only on the demand-driven changes in current loan portfolio after controlling for company-level, department-level fixed characteristics as well as differences in loan portfolio values due to monthly seasonality and yearly growth. This results support the argument that farmers use microcredit borrowings as a consumption smoothing tool in times of adverse weather events. The results of the estimation based on urban loan portfolio confirm that rainfall shocks have no significant impact on the current urban loan portfolio supporting the va-lidity of results (see Table B4 in Appendix B for details).

The results in Table 13 show that negative as well as positive rainfall shocks are associated with larger overdue portfolio taking into account monthly and yearly trends.

4 _{The increase of overdue portfolio in response to negative rainfall shocks (droughts)}

support the Ability hypothesis, meaning that in times of droughts farmers do not have enough income to maintain stable loan repayment. The point estimate of SHOCK NEG shows that in case of negative rainfall shocks the monthly per-department overdue rural loan portfolio is on average larger by around 34 thousand Boliviano.

4_{We perform System GMM estimation to check the robustness of results (see Tables B1-B3 in Appendix}

Table 13: Fixed effects estimation results: Effect of weather shocks on the overdue rural loan portfolio

This table summarizes the results of fixed effect estimation of rainfall shocks and the overdue rural loan portfolio. For each variable the coefficient and its standard error are presented. Model 2.1 and 2.2 estimate the impact of the absolute rainfall level on the overdue rural loan portfolio. Models 2.3 and 2.4 estimate the impact of negative and positive rainfall shocks.

Model 2.1 Model 2.2 Model 2.3 Model 2.4 RAIN -17,393.7*** -8,617.5* (4,091.8) (4,677.0) L.RAIN 6,769.6*** -2,002.4 (2,391.7) (3,180.5) L2.RAIN -343.2 -4,028.6 (4,104.6) (4,091.3) SHOCK NEG 20,164.6** 33,444.7* (9,694.4) (17,950.0) L.SHOCK NEG 2,024.5 23,265.4 (13,923.9) (22,079.2) L2.SHOCK NEG 51,102.1*** 39,327.7** (16,358.3) (18,047.0) SHOCK POS -25,187.5* 11,489.9 (13,702.7) (9,715.2) L.SHOCK POS 445.9 24,953.1* (17,903.1) (14,582.6) L2.SHOCK POS -7,065.1 21,554.5* (12,269.1) (12,372.9) Constant 346,586.3*** 175,265.6*** 314,203.5*** 62,173.3 (14,273.7) (55,231.6) (8,695.1) (62,431.4) Observations 1,655 1,655 1,655 1,655 R-squared 0.0 0.1 0.0 0.1 No. of IDs 58 58 58 58

MFI FEs Yes Yes Yes Yes

Province FEs Yes Yes Yes Yes

Time FEs No Yes No Yes

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

support both the Mobility and Ability hypotheses. Poor repayment performance of loans may be related to the damaged infrastructure, as, for example, flooded roads in rural areas may create difficulties for rural clients to reach MFI branches for repayment. In the same time, excessive rainfall in the harvest period bears significant risks for farmers. Pelka et al. (2015) documents that excessive amount of precipitation in the harvesting period of rice increases the credit risk of loans granted to small-scale farmers in Madagascar. In this case, the negative impact of excessive rainfall shocks on repayment performance may support the Ability hypothesis. As long as we do not have the necessary data on crops cultivated in different regions in Bolivia, we cannot control for their specific harvesting periods and disaggregate the ability and mobility channels through which excessive rains damage repayment performance. The placebo estimation based on urban loan portfolio suggests no evidence of significant impact of rainfall shocks on repayment performance of urban loans supporting the validity of results (see Table B5 in Appendix B for details).

Table 14: Fixed effects estimation results: Effect of weather shocks on the share of overdue rural loan portfolio

This table summarizes the results of fixed effect estimation of rainfall shocks and the share overdue rural loan portfolio. For each variable the coefficient and its standard error are presented. Model 3.1 and 3.2 estimate the impact of the absolute rainfall level on the share of overdue rural loan portfolio. Models 3.3 and 3.4 estimate the impact of negative and positive rainfall shocks.

Model 3.1 Model 3.2 Model 3.3 Model 3.4 RAIN 0.207 0.360 (0.163) (0.267) L.RAIN 0.241** 0.178 (0.112) (0.178) L2.RAIN 0.203* 0.348* (0.103) (0.192) SHOCK NEG -0.316 -0.115 (0.390) (0.519) L.SHOCK NEG -0.613 -0.230 (0.407) (0.441) L2.SHOCK NEG -0.265 -0.459 (0.382) (0.496) SHOCK POS 0.516 0.660 (0.520) (0.604) L.SHOCK POS -0.206 -0.395 (0.472) (0.617) L2.SHOCK POS 0.367 0.265 (0.341) (0.457) Constant 0.267 -2.148 2.144*** 2.946 (0.966) (2.558) (0.294) (2.087) Observations 1,536 1,536 1,536 1,536 R-squared 0.036 0.051 0.004 0.036 No. of IDs 56 56 56 56

MFI FEs Yes Yes Yes Yes

Province FEs Yes Yes Yes Yes

Time FE No Yes No Yes

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

new loan disbursements. In order to ensure that the instruments in the model are valid, we test for their exogeneity and relevance. To test whether the instruments are uncor-related with the error term, we apply the Sargan test of overidentifying restrictions. We can not reject the null hypothesis of the exogeneity of instruments in Models 4.2 and 4.4, where time fixed effects are controlled for.

Table 15: 2SLS estimation results: Effect of rural loan portfolio performance on the amount of newly disbursed loans

This table summarizes the results of 2SLS estimation of overdue loans and new loan disbursements. For each variable the coefficient and its standard error are presented. Models 4.1 and 4.2 estimate the impact of lagged overdue loan portfolio on the new loan disbursements with pooled 2SLS estimation while Models 4.3 nd 4.4 present the results of panel 2SLS estimation. Overdue loans are instrumented by negative rainfall shocks.

Pooled 2SLS Panel 2SLS Model 4.1 Model 4.2 Model 4.3 Model 4.4 L.OVERDUE RUR 4.302** 4.211** 6.298** 10.120*** (1.998) (2.131) (3.069) (4.608) Constant -959,589 -318,958 (625,629) (518,196) Observations 1,644 1,644 1,644 1,644 Number of ID 58 58

Time FEs No Yes No Yes

Post-estimation tests Exogeneity of instruments

Sargan test statistic 3.523** 0.679 3.033** 1.061 Relevance of instruments

F-test statistic 4.8*** 4.8*** 4.8*** 4.8*** Endogeneity of L.OVERDUE RUR

Durbin-Wu-Hausman test statistic 3.523** 5.862*** 10.425*** 15.325*** Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

instru-mental variable is equal to zero in the first-stage regression. We reject the null hypothesis of irrelevant instruments in all specifications, though the F-statistics are not very large. Based on the exogeneity and relevance tests, we conclude that the instruments are valid. After the estimation is conducted, we test the endogeneity of the independent variable. For all specifications, we reject the null hypothesis of the Durbin-Wu-Hausman test stating that the presumed endogenous variable under consideration can be treated as exogenous. Although the estimation results point to a positive association between loan portfolio performance and future loan disbursements, conclusions regarding the supply-side effects should be made with cautiousness. We observe the amount of new loan disbursements in equilibrium where demand and supply of new loans are simultaneously determined. Thus, the positive impact of overdue loans can, on the one hand, be explained by the supportive behavior of MFIs in the events of weather shocks when they provide additional funds to help borrowers overcome negative impacts of shocks. On the other hand, borrowers themselves may apply for new loans to recover from disasters.

Summarizing, the results of the analysis provide evidence in support of the Demand hypothesis and show that in rural areas farmers may increase borrowings in times of ad-verse weather events. The results also suggest that rainfall shocks are associated with larger overdue loans. At this point, however, we are not able to distinguish between mobility and ability channels. We finally observe some evidence on the relation between loan repayment performance and future loan disbursements. We are not, however, able to either accept or reject the Supply hypothesis, as long as the simultaneous determination of supply and demand of loans does not allow us to disentangle supply-side effects.

6

Discussion and Conclusion

weather shocks among poor rural households. The availability of microcredit programs positively affects the ability of households to cope with disasters reducing their vulnera-bility to poverty. There is, however, no evidence on the impact of weather shocks on the aggregate demand of microcredit. Our results suggest evidence in favor of this argument and show that higher rainfall levels in absolute terms are associated with smaller current rural loan portfolio. Negative rainfall shocks (such as droughts) measured as below -1 standard deviation of rainfall level from long-term mean values are associated with larger monthly per-department current rural loan portfolios of MFIs. We do not observe evi-dence in favor of the alternative hypothesis discussed earlier suggesting that poor rural households may postpone large investments in businesses and other expenditures in times of adverse weather shocks thus reducing the demand for credit. The current rural loan portfolio used in the analysis, however, includes loans of different sizes. The effect of weather shocks on credit demand may be heterogeneous based on the size of loans as larger loans are more likely to be used for investment purposes. Examination of possible heterogeneous effects based on the loan size may be a subject for further research.

The current rural portfolio used in the analysis, however, includes loans targeted to other activities as well such as retail and wholesale. The impact of weather shocks on the repay-ment performance most likely varies with the purpose of loans. Identification of possible heterogeneous effects based on the purpose of loans is a subject for a future study.

Literature does not yet provide a thorough analysis of supply-side constraints re-sulting from weather-related shocks. We make an attempt to investigate the reaction of NGOs in Bolivia to their loan portfolio deterioration as a result of adverse rainfall shocks. We provide some evidence on the positive relation between overdue loans and the amount of new loan disbursements. One of the limitations of the analysis is that the growth of current loan portfolio is not a perfect measure for new loan disbursements. As long as current loan portfolio measures the amount of interest-bearing outstanding loans, its growth not only reflects the amount of newly issued loans, but also the amount of old loans that have become overdue between 2 periods, at the same time not reflecting the amount of new loans that became overdue until the end of the second period. Another concern relates to the quality of the instruments. Although we show that rainfall shocks have a significant impact on the amount of overdue loan portfolio, the F-statistic in the first stage regressions is not very large indicating a low explanatory power of the instruments. Another important limitation of the analysis is that new loans are observed in equilibrium where supply and demand are simultaneously determined. The positive impact of over-due loans can be explained by the supportive behavior of MFIs in the events of weather shocks when they provide additional funds to help borrowers overcome negative impacts of shocks. This type of behavior was observed after the 1998 floods in Bangladesh. On the other hand, borrowers themselves may apply for new loans to recover from disasters. Thus, we are not able to separate supply-side effects, though the results give some indi-cation of the relation between repayment performance and new loan disbursements.

among 9 departments. Although we use the average rainfall levels based on the number of coordinates in each department, the resulting values do not perfectly reflect the weather conditions in a given department. A more detailed data on rainfall covering larger areas could potentially improve the results of the analysis.

References

Ahmed, Syud A., Noah S. Diffenbaugh, and Thomas W. Hertel, “Climate Volatility Deepens Poverty Vulnerability in Developing Countries,” Environmental Research Letters, 2009, 4, 1–8.

Barnett, Barry J. and Olivier Mahul, “Weather Index Insurance for Agriculture and Rural Areas in Lower-Income Countries,” American Journal of Agricultural Economics, 2007, 89 (5), 1241–1247.

Blundell, Richard and Stephen Bond, “Initial Conditions and Moment Restrictions in Dynamic Panel Data Models,” Journal of Econometrics, 1998, 87 (1), 115–143.

Collier, Benjamin L., “Natural Disasters and Credit Supply Shocks,” Wharton Risk Management Center, University of Pennsylvania. Working Paper #2014-02, 2014, pp. 1–31.

and Jerry R. Skees, “Increasing the Resilience of Financial Intermediaries through Portfolio Level Insurance against Natural Disasters,” Natural Hazards, 2012, 64 (1), 55–72.

, Ani L. Katchova, and Jerry R. Skees, “Loan Portfolio Performance and El Ni˜no Events, an Intervention Analysis,” Agricultural Finance Review, 2011, 71, 98–119.

, Mario J. Miranda, and Jerry R. Skees, “Natural Disasters and Credit Supply Shocks in De-veloping and Emerging Economies,” Wharton Risk Management Center, University of Pennsylvania. Working Paper #2013-03, 2013, pp. 1–19.

Del Ninno, Carlo, Paul A. Dorosh, and Lisa C. Smith, “Public Policy, Markets and Household Coping Strategies in Bangladesh: Avoiding a Food Security Crisis Following the 1998 Floods,” World Development, 2003, 31 (7), 1221–1238.

Demont, Timothee, “Poverty, Access to Credit and Absorption of Weather Shocks: Evidence from Indian Self-Help Groups,” CRED working paper, 2013, (214705).

Dercon, Stefan and Pramila Krishman, “In Sickness and in Health : Risk Sharing within Households in Rural Ethiopia Ethiopia Pramila Krishnan,” Journal of Political Ecoomy, 2000, 108 (4), 688–727.

Dowla, Asif, “Climate Change and Microfinance,” Washington: Grameen Fundation, 2009.

FAO, “The state of Food and Agriculture: Climate Change, Agriculture and Food Security,” Food and Agriculture Organization of the United Nations, 2016, pp. 1–166.

Fichera, E and D Savage, “Income and Health in Tanzania. An Instrumental Variable Approach,” World Development, 2015, 66 (June), 500–515.

Gertler, Paul, David I. Levine, and Enrico Moretti, “Do Microfinance Programs Help Families Insure Consumption Against Illness?,” Health economics, 2009, 18 (3), 257–273.

Hammill, Anne, Richard Matthew, and Elissa Mccarter, “Microfinance and Climate Change Adaptation,” IDS Bulletin, 2008, 39 (4), 113–122.

IPCC, “Climate Change: Fourth Assessment Report.,” Intergovernmental Panel on Climate Change (IPCC), Geneva, Switzerland, 2007, pp. 1–104.

Islam, Asadul and Pushkar Maitra, “Health Shocks and Consumption Smoothing in Rural House-holds: Does Microcredit have a Role to Play?,” Journal of Development Economics, 2012, 97 (2), 232–243.

Kaboski, Joseph P and Robert M Townsend, “Policies and Impact: An Analysis of Villagelevel Microfinance Institutions,” Journal of the European Economic Association, 2005, 3 (1), 1–50.

Kazianga, Harounan and Christopher Udry, “Consumption Smoothing? Livestock, insurance and Drought in Rural Burkina Faso,” Journal of Development Economics, 2006, 79, 413–446.

Khandker, Shahidur R., “Coping with Food: Role of Institutions in Bangladesh,” Agricultural Eco-nomics, 2007, 36 (2), 169–180.

Lensink, Robert, Roselia Servin, and Marrit van den Berg, “Do Savings and Credit Institutions Reduce Vulnerability? New Evidence From Mexico,” Review of Income and Wealth, 2017, 63 (2), 335–352.

Library of Congress Federal Research Division, “Country Profile: Bolivia,” 2006.

Maccini, Sharon and Dean Yang, “Under the Weather: Health, Schooling, and Economic Conse-quences of Early-Life Rainfall,” The American Economic Review, 2009, 99 (3), 1006–1026.

Mehar, Mamta, Surabhi Mittal, and Narayan Prasad, “Farmers Coping Strategies for Climate Shock: Is It Differentiated by Gender ?,” Journal of Rural Studies, 2016, 44, 123–131.

Miguel, Edward, Shanker Satyanath, and Ernest Sergenti, “Economic Shocks and Civil Conflict: An Instrumental Variables Approach,” Journal of Political Economy, 2004, 112 (4), 725–753.

Morduch, Jonathan, “Poverty and Vulnerability,” The American Economic Review, 1994, 84 (2), 221–225.

Moser, Rafael M. B. and Lauro Gonzalez, “Microfinance and Climate Change impacts: The Case of Agroamigo in Brazil,” RAE - Revista de Administra¸c˜ao de Empresas, 2015, 55 (4), 397–407.

Pelka, Niels, Ron Weber, and Oliver Musshoff, “Does Weather Matter ? How Rainfall Shocks affect Credit Risk in Agricultural Microfinance,” in “International Conference of Agricultural Economics” 2015, pp. 1–23.

Pitt, Mark M. and Shahidur R. Khandker, “Credit Programmes for the Poor and Seasonality in Rural Bangladesh,” The Journal of Development Studies, 2002, 39 (2), 1–24.

Roodman, David, “How to Do xtabond2: An Introduction to ”Difference” and ”System” GMM in Stata,” 2006.

Rosenzweig, Mark R. and Hans P. Blinswanger, “Wealth, Weather Risk and the Composition and Profitability of Agricultural Investments,” The Economic Journal, 1993, 103 (416), 56–78.

Shah, Shekhar, “Coping with Natural Disasters: The 1998 Floods in Bangladesh,” in “Seminar Paper Presented to the World Bank Summer Research Workshop on Poverty and Development” 1999.

Stern, Nicholas, “The Economics of Climate Change,” The Stern Review. Cambridge: Cambridge University Press, 2007.

Thurlow, James, Tingju Zhu, and Xinshen Diao, “The Impact of Climate Variability and Change on Economic Growth and Poverty in Zambia,” International Food Policy Research Institute, 2009. Trivelli, Carolina, “Rural Finance and Insurance on the North Coast of Peru. Summary Report 2005/06.

Institudo de Estudios Peruanos,” Technical Report 2006.

Udry, Christopher, “Risk and Insurance in a Rural Credit Market: An Empirical Investigation in Northern Nigeria,” The Review of Economic Studies, 1994, 61 (3), 495–526.

Appendix

Appendix A: Data Description

Table A1: Descriptive statistics of urban loan portfolios based on repayment status

This table presents the mean, standard deviation, minimum and maximum values of urban loan portfolios of FINRURAL member NGOs over the period Oct.2013-Jul.2016 disaggregated by the repayment status.

Variable Mean Std.Dev. Min Max

Gross Urban Loan Portfolio 177.00 183.00 6.64 571.00 Current Urban Loan Portfolio 173.00 182.00 6.27 562.00 Restructured Current Urban Loan Portfolio 0.61 0.95 0.00 5.23 Overdue Urban Loan Portfolio 2.20 1.71 0.00 7.00 Restructured Overdue Urban Loan Portfolio 0.09 0.15 0.00 0.84 Progressive Urban Loan Portfolio 0.95 1.15 0.00 4.76 Restructured Progressive Urban Loan Portfolio 0.05 0.14 0.00 0.83 Contingent Urban Loan Portfolio 0.00 0.00 0.00 0.00

Source: Author’s calculations based on FINRURAL data.

Table A2: Summary of urban loan portfolio quality in 9 departments in Bolivia

This table presents the means and standard deviations of 3 portfolio quality ratios in 9 departments in Bolivia: share of overdue urban loan portfolio in gross urban loan portfolio, share of restructured urban loan portfolio in gross urban loan portfolio and share of restructured overdue urban loan portfolio in gross urban loan portfolio

Department Share of Overdue Share of Restructured Share of Restructured Urban Loan Urban Loan Overdue Urban Portfolio Portfolio Loan Portfolio

Beni 1.242 0.025 0.000 (1.146) (0.080) (0.000) Chuquisaca 0.973 0.055 0.001 (0.963) (0.123) (0.008) Cochabamba 6.024 0.605 0.101 (15.156) (0.901) (0.148) La Paz 1.670 1.705 0.237 (1.541) (2.482) (0.445) Oruro 1.125 1.171 0.008 (1.313) (0.307) (0.044) Pando 2.051 0.595 0.135 (2.175) (0.890) (0.237) Potosi 3.165 0.056 0.034 (4.821) (0.095) (0.084) Santa Cruz 1.698 0.700 0.215 (2.674) (1.047) (0.32) Tarija 2.735 0.957 0.132 (3.149) (0.960) (0.205)

Table A3: Descriptive statistics of urban loan portfolios based on loan sizes

This table summarizes the urban loan portfolios of FINRURAL member NGOs over the period Oct.2013-Jul.2016 disaggregated by the loan sizes. Columns 2 and 3 present the means and standard deviations of urban loan portfolio while columns 4 and 5 present means and standard deviations of number of clients in different loan size ranges.

Loan Portfolio Number of clients

Range Mean Std.Dev. Mean Std.Dev

Greater than USD 200,001 0.1 0.57 0.06 0.33 USD 100,001 - USD 200,000 1.90 5.4 2.46 6.20 USD 50,001 - USD 100,000 5.123 9.07 11.54 18.98 USD 30,001 - USD 50,000 6.51 9.60 28.88 64.68 USD 20,001 - USD 30,000 6.06 9.32 46.66 77.57 USD 15,001 - USD 20,000 7.38 10.60 100.03 180.52 USD 10,001 - USD 15,000 14.60 20.10 400.66 696.31 USD 5,001 - USD 10,000 39.80 50.00 1,749.78 2,806.25 USD 4,001 - USD 5,000 12.70 15.10 799.80 1,151.97 USD 3,001 - USD 4,000 13.40 13.00 1,090.89 1,413.92 USD 2,001 - USD 3,000 23.00 24.30 2,191.79 2,454.76 USD 1,001 - USD 2,000 29.90 33.10 4,500.70 5,287.16 USD 501 - USD 1,000 24.40 35.40 6,803.00 10,199.11 Smaller than USD 500 15.40 24.40 9,640.49 14,163.80

Source: Author’s calculations based on FINRURAL data.

Table A4: Descriptive statistics of urban loan portfolios based on targeted sectors

This table presents the mean, standard deviation, minimum and maximum values of urban loan portfolios of FINRURAL member NGOs over the period Oct.2013-Jul.2016 disaggregated by the targeted sectors. Rows 1 and 2 summarize the loan portfolio of productive and non-productive sectors while rows 3 and 4 summarize the number of operations.

Variable Mean Std.Dev. Min Max

Total Urban Loan Portfolio 52.30 66.60 1.87 2253.00 to Productive Sectors

Total Urban Loan Portfolio 148.00 144.00 5.23 498.00 to Non-Productive Sectors

Total Number of Urban Operations 2,841.09 3,992.34 68.00 15,862 in Productive Sectors

Total Number of Urban Operations 21,722.15 3,992.34 151 138,631 in Non-Productive Sectors

Source: Author’s calculations based on FINRURAL data.

Note: Loan portfolio values are presented in millions Boliviano (rows 1 and 2)