Climate Change and Household Debt in Rural India

Abstract

Climate change and indebtedness have been repeatedly highlighted as major causes of distress for rural households in India. However, despite the close connection between climate conditions and rural livelihoods, there has been little attempt to systematically examine the association between the two. To address this gap, we combine national-level longitudinal data from IHDS, MERRA-2, and the Indian Ministry of Agriculture to study the impact of climate anomalies on household indebtedness across rural India. Using a longitudinal approach that accounts for potential confounders at household, village, and district levels, we find pervasive effects of season-specific, five-year climate anomalies on multiple dimensions of household debt, particularly in arid and semi-arid areas. Most notably, temperature anomalies in the winter cropping season in arid and semi-arid areas are associated with increasing household indebtedness. We further find that climate change interacts with existing socioeconomic differences—caste and landholding in particular—to deepen both the size and the depth of indebtedness for rural households.

Introduction

Climate change and indebtedness have been repeatedly highlighted as two of the most important factors behind the heightened vulnerability of rural households in India (Dasgupta, et.al, 2014; Ramprasad, 2018; Taylor, 2018). Various dimensions of climate change such as shifts in rainfall, an increasing incidence of droughts and floods, and a rise in daily mean temperatures are already undermining the productivity of several key South Asian crops including wheat, rice, and maize which are central to the rural economy (Davis, et.al, 2019; Oxfam, 2019; Palanisami, et.al, 2019). These shifts have been shown to have adverse impacts on the labor force participation of rural households (Lee and Lim, 2018; Lundgren-Kownacki, et.al., 2018) and to widen the gender wage gap (Mahajan, 2017), further undermining the livelihoods of the vulnerable.

At the same time, rising indebtedness among rural households, especially farmers, is claimed to have forced millions out of agriculture and pushed thousands towards distress migration, and, in extreme cases, suicide (Ghosh, 2013; Taylor, 2013; Vaditya, 2017). Debt itself is not necessarily damaging and, indeed, is an important instrument with which households achieve consumption smoothing (Garikipati, 2017) and navigate negative events such as crop failure and health exigencies (Collins, et.al 2009). However, the social, economic, and ecological conditions that frame debt transactions can make them onerous for borrowing households, placing them under tremendous pressure (Gerber, 2013; Guerin, 2014; Ramprasad, 2018). Social and economic marginalization, erosion of natural resources like land and water, and poorly designed state policies can render otherwise small sums of debt extremely burdensome (Bhaduri, 1986; Gerber, 2013 and 2014; Kumar and Venkatachalam, 2018).

Climate change is expected to further undermine the financial wellbeing of rural households (Dasgupta, et.al, 2014). Climate variability has been shown to depress household revenues (Kar and Das, 2015) and crop productivity (Arora and Birwal, 2017), both proximate causes of the increase in rural household indebtedness (Mukherjee, 2009). However, very few studies have systematically examined the links between climatic shifts and household debt. Our paper addresses this gap in literature by asking: How do climate anomalies affect the indebtedness of rural households in India, and how do social and ecological conditions modify this relationship? We use longitudinal household survey data from the India Human Development Survey together with climate data from NASA and agroecological data from the Indian Ministry of Agriculture to examine how household indebtedness evolved from 2005–2011 under varying climate exposures, measured as season-specific five-year climate anomalies and village-level shocks. We find that that in arid and semi-arid regions, temperature anomalies in the winter cropping season (October-February) and village-level droughts have large positive effects on both the incidence and depth of household indebtedness.

Previous Research

Climate change in India

The fifth Assessment Report (AR-5) of the Intergovernmental Panel on Climate Change (IPCC) estimated it very likely that the mean annual temperature has increased over the past century in South Asia and the number of warm days has increased since 1950 (Hijioka, et.al, 2014; pp. 1333). The report further noted the declining trend in rainfall, with the number of deficit monsoons increasing. AR-5 projected a significant increase in temperature and precipitation extremes across South Asia.

Scholars argue that in a large and diverse country like India, climatic shifts are likely to vary spatially and seasonally (Alexandru and Sushama, 2015; Palanisami, et.al, 2019; Sarthi 2019). For instance, the increase in extreme warm days, was found to bemore persistent in the eastern Himalayas, the southern plateau and the Eastern and Western Ghats, while the lower Gangetic plains and the eastern plateau region witnessed an increase in the number of extreme warm days only during the months of January, August, September, and November (Palanisami, et.al, 2019). Shifts in rainfall are similarly differentiated. In both northeastern and southeastern regions of India, rainfall during the January-March period is expected to decline significantly while the quantum of rainfall received during the months of July-September is expected to increase (Palanisami, et.al, 2019). Sarthi (2019) additionally finds that, while regions like the Gangetic plains and the Western Ghats are likely to witness excess rainfall during the monsoonal period, large parts of western and Peninsular India are likely to experience rainfall deficits.

Given these spatial and seasonal dimensions of climate change, the consequences of these shifts for rural livelihoods are also likely to play out spatially and seasonally. This is especially so in rural India because of the fact that livelihood activities and their outcomes there are profoundly shaped by seasons and by the agroecological particularities of a given region (Ranaware, et.al, 2015; Nagaraj, et.al, 2016; Ramdas, 2018). By examining how indebtedness is differentiated across regions and over different agricultural seasons, we account for the spatial and temporal dimensions of climate change.

Climate change and rural livelihoods

Approximately 70% of the Indian population lives in rural areas (Census of India, 2011). Agriculture and allied activities are the mainstay of the rural economy and employ 59% of the national workforce (Himanshu and Kundu, 2016; Basole, 2017). A majority of Indian farmers specifically depend on rainfed agriculture (Singh, et.al, 2017; Davis, et.al, 2019; Zaveri and Lobell, 2019), and, while there has been an increase in the area under irrigation, most of this irrigation comes from groundwater, which, in regions like peninsular India, often depends on annual rainfall for recharge (Kulkarni, 2011; Meinzen-Dick, et.al, 2018; Stratton, et.al, 2019). The bulk of food in India is produced in two agricultural seasons: the monsoon or kharif season (June-September), underpinned by the southwest monsoon, and the winter or rabi season (October-January) that relies on the northeast monsoon (Nalinakanthi, 2018). Rice, largely grown in the monsoon season, and wheat, grown in the winter, account for 40% and 37% respectively of the total grain production in India (Min. of Agriculture, 2018; Nalinakanthi, 2018).

Both rice and wheat are susceptible to variations in temperature and precipitation (DeFries, et.al, 2016; Davis, et.al, 2019). Rice yields are vulnerable to extreme warm days (Chakraborty and Kumar, 2018) and to interannual temperature and rainfall variations (Davis, et.al, 2019). It is estimated that roughly 40% of the rice-growing locations in India could become unsuitable due to shifts in temperature (Singh, et.al, 2017). Wheat yields, similarly, are already declining due to rising temperatures and will continue to do so (Padakandla, 2016; Zaveri and Lobell, 2019), particularly due to an increase in the occurrence of extreme warm days during the winter cropping season (Chakraborty and Kumar, 2018). In addition to temperature, untimely rainfall can also lead to crop loss. Untimely rainfall has also been shown to trigger floods, whether in the Himalayan states of Uttarakhand and Himachal Pradesh or in the North-Eastern states like Assam and Meghalaya (DTE, 2020; Sarkar, 2020), which can in turn result in significant crop losses.

Climate change can also affect rural livelihoods indirectly. One such pathway is via health. Children exposed to extreme weather events in utero have been shown to be stunted and underweight (Shively, et.al, 2015; Kumar, et.al, 2016; Shively, 2017). Climatic exposures can also affect the health of adults. Excess heat can increase the risk of heatstroke and cardiovascular stress, particularly for people involved in manual labor (Thomas and Jagil, 2021; Sahu, et.al, 2013; Crowley, 2016; Lundgren-Kownacki, 2018). Similarly, the risk of waterborne diseases like cholera and typhoid as well as vector-borne diseases like malaria increases with temperature (Dhara, et.al, 2013). In India the incidence of waterborne and respiratory illnesses increases during the monsoon season, with a second spike occurring during the winter season (Narayan, et.al, 2020). Increases in the incidence of health ailments translates into a corresponding increase in healthcare expenditures, which can in turn cripple rural households economically. Poor health can also undermine participation in the labour force.

Finally, climate change can impact rural livelihoods is by increasing the rate of rural out-migration (Surie and Sharma, 2019), and by depressing rural wage rates (Mahajan, 2017).

Rural indebtedness

While debt conjures up images of suffering and exploitation, we must also recognize that it is an important tool for households to cope with irregular income flows or shocks and achieve consumption-smoothing (Collins, et.al, 2009; Taylor, 2012; Rana and Vishwanathan, 2019). Among rural households, especially those who depend on agriculture and allied activities, income flows are often uncertain (Collins, et.al, 2009) and, in this context, taking on debt is a way to meet agricultural input costs, everyday consumption needs, and various unforeseen expenses. Both formal and informal sources play an important role in this process. While loans from the former come with a lower interest rate, those from the latter come with lesser paperwork and fewer formalities. Indeed, local moneylenders, much maligned by scholars (Ghosh, 2013; Vaditya, 2017) are key players in the rural debt markets, extending timely credit with minimal paperwork and procedures (Sethi, 2018; Guerin and Venkatasubramanian, 2020).

However, while debt is vital for consumption smoothing, an excess of debt has been shown to be a major cause for distress among rural households (Ghosh, 2013; Dandekar and Bhattacharya, 2017). The term overindebtedness has been used in both colloquial and academic contexts to describe a situation where the repayment of debt becomes a major financial burden for a household (Chichaibelu and Waibel, 2017) and where the borrower makes sacrifices to meet the repayment deadlines (Schicks, 2014). Poorer households have been observed to borrow large sums of money relative to their incomes and therefore become overindebted (Chichaibelu and Waibel, 2018). In India, indebtedness has been on the rise among rural households (Taylor, 2013; NSSO, 2014; Dandekar, 2017). Further, the magnitude of indebtedness is differentiated by class (Kandikuppa, 2021), gender (Ghosh, 2013), agricultural practices (Green, 2021), and by the policy environment (Taylor, 2013 and 2018).

Various agronomic, infrastructural, economic, and socio-cultural factors appear to have contributed to the trend of rising indebtedness among rural households. An increase in the cost of cultivation has been one of the key drivers (Ghosh, 2005 & 2013; Taylor, 2013; Vaditya, 2017). Stagnant yields and falling farm incomes have pushed many farmers to adopt intensive agricultural practices marked by heavy reliance on groundwater abstraction, purchased seeds, and chemical fertilizers and pesticides (Ghosh, 2013; Taylor, 2013; Vaditya, 2017). Such practices have increased the cost of cultivation, allowed industrial capital to make inroads into the Indian countryside, and increased indebtedness (Gupta, 1998; Ghosh, 2005; Vaditya, 2018; Aga, 2019). Among agricultural workers, rural wage rates have been either stagnant or have grown only marginally since 2010 (Himanshu and Kundu, 2016; Basole, 2017)

One of the most common reasons for borrowing in rural India is healthcare costs. Unforeseen health emergencies or prolonged healthcare needs force many rural households to borrow heavily to meet the attendant costs (Pradhan, 2013; NSSO, 2014; Krishna, 2017), reflecting a lack of affordable and accessible healthcare in rural India. Social ceremonies are another major driver of indebtedness among rural households who are often forced to borrow to meet the expenses of death feasts, weddings, funerals and other ceremonies (Krishna, 2006 & 2010; Banerjee and Duflo, 2016; Dandekar and Bhattacharya, 2017).

Another key driver of rural household indebtedness is credit rationing. Imperfect market conditions due to asymmetry of information between lenders and borrowers lead to situations where formal lending institutions are unwilling to loan the desired amount to borrowers (Asante-addo, et.al, 2016; Das and Laha, 2017). In some instances, the formal institutions do not have full information about the potential borrowers, and, in others, the borrowers may find the procedures to get a loan from a formal institution to be challenging or may feel that their loan application would be turned down (Kochar, 1997; Asante-addo, et.al, 2016; Das and Laha, 2017). In either case, the credit needs of the borrowers are not fully met by the formal institutions forcing the former to approach informal moneylenders, who tend to charge higher interest on the loans.

Beyond these proximate drivers of indebtedness, the social structures underpinning rural society are equally or more responsible for debt-related distress among rural households. Semi-feudal social structures marked by unequal power relations between farmers and wage laborers on the one hand and moneylenders on the other persist in rural areas, alongside forced commerce wherein farmers and agricultural wage workers are forced to borrow at high interest rates and exchange their produce or labour to the moneylenders in lieu of the repayment (Bell, 1976 & 1977; Bhaduri, 1973 & 1986; Dhanagare, 2016). The identity of both the borrowers and lenders also shapes the former’s experiences of living with debt. Oftentimes, the borrowers’ caste determines the terms on which lenders extend credit (Kumar and Venkatachalam, 2018). Further, the uncertainties inherent to agriculture, state policies like loan waivers, the emergence of new sources of credit such as banks and microfinance institutions alongside the persistence of traditional ones like informal moneylenders all contribute to make debts exploitative (Taylor, 2013; Carswell, 2020; Green, 2020). Another reason why debts are exploitative especially for the rural poor including marginal and small farmers and landless laborers is that these households are dependent wealthier households like large landholders, salaried people, and agricultural commodity traders who control most of the means of production (Bhaduri, 1986; Hardiman, 1996). Since the rural poor usually control nothing more than small landholdings and their own labor, they are forced to rely on borrowings for all other means of production, which in turn traps them into exploitative debt relations.

The Indian state has striven to address rural indebtedness first by increasing credit supply through programs like the Kisan Credit Card Scheme and the more recent Pradhan Mantri Jan Dhan Yojana and, from to time, by writing off loans owed by farmers to government owned (public sector) banks. However, as has been shown (Guerin and Subramanian, 2020; Kandikuppa, 2021), despite these initiatives informal moneylending has thrived and has indeed enjoyed resurgence in recent decades (Kandikuppa, 2021). This is because despite the steep interest rates, these lending sources have lesser paperwork and make loans available on short notice. Nonetheless, reliance on informal sources has been shown to exacerbate the debt burdens of rural households.

Debt can also add to the vulnerability of rural households by forcing them to reorganize their lives and livelihoods in order to generate the surplus needed for repayment (Gerber, 2013 & 2014). Such reorganization often includes the adoption of intensive agricultural practices (Ghosh, 2005 & 2013; Taylor, 2012 & 2013; Vaditya, 2017), undertaking distress migration (Breman, 1996; Mosse, et.al, 2002), and borrowing further to repay earlier loans (Waibel and Chichaibelu, 2017). Debts also force households to make various sacrifices, including of essentials like food and medication (Drentea, 2000; Drentea and Lavrakas, 2000; Schicks, 2014). Indebtedness also has severe mental health outcomes for the borrowers (Drentea, 2000) and has been linked to thousands of farmer suicides (Gill and Singh, 2000; Ghosh, 2005; Swaminathan, 2006; Vaditya, 2017).

Debt and climate change

Both climate change and debt add to the “hazardscapes” (Ribot, 2014) of rural households. The AR-5 anticipates, with ‘high confidence’, that climate change will significantly affect rural economic activities, livelihoods and land use (Dasgupta, et.al, 2014) For instance, an increase in temperature during January, April, and July have been shown to result in a loss of household revenue (Kar and Das, 2015); and these losses are more pronounced for small farmer and marginal farmers. Loss of crops and income can push farmers to increase borrowing, to meet their consumption needs or to make provisions for future investments. At the same time, the prospect of such losses can prompt farmers to take adaptive measures to climate-proof agriculture, including investing in tube-wells, improved seeds, or other agronomic practices. In both these scenarios, rural households, particularly farmers, are likely to increase their borrowings. Our paper empirically establishes the relationship between climate change and household debt by showing how changes in temperature, rainfall, and occurrence of extreme events impact the latter.

Data and Methods

Data

We combine data from the India Human Development Survey, NASA’s Modern Era Representative Analysis for Research and Applications-2 (MERRA-2), and the Indian Ministry of Agriculture to examine how anomalies in rainfall and temperature impact both the size and depth of household indebtedness (Darling, 1928; NSSO, 2014). Household-level data is derived from the Indian Human Development Survey-I and II (IHDS-I & IHDS-II). IHDS is a multi-topic panel dataset resulting from two survey waves conducted in 2004–05 and 2011–12. It is the only comprehensive, publicly-available, nationally-representative panel dataset from India. IHDS comprises a sample of 41,554 households drawn in 2004–05, with 83% of the households being retained in 2011–12 (Desai and Vanneman, 2015). Our analysis includes 21,485 rural households that were surveyed in both rounds of IHDS and that had borrowed at least once in either 2005 and 2011, representing 1420 villages, 275 districts and 33 states. To prevent the aging of the sample, we include local split-off households that were formed by 2004–05 household members during the inter-survey period and were then interviewed by IHDS in 2011–12. We focus on the rural sample which is particularly exposed to both climate change and indebtedness and for which village-level data are available.

From this source, we extract household-level data on socioeconomic characteristics including primary occupation, caste, landholdings, and gender of the household head, as well as information on debt and resources including the total number of outstanding loans, the amount owed, total consumption expenditure, and gross income (Table 1).

Table 1.

Descriptive statistics for the dependent and independent variables at household and village levels.

Dependent Variables	Arid Districts (%)	Humid Districts (%)
Increase in loans	49.8	52.6	No. loans borrowed increased between 2005 & 2011
Increase in debts	50.9	51.7	Loan amount increased between 2005 & 2011
Increase in DIR	50.0	54.3	DIR increased between 2005 & 2011
Increase in DCR	47.9	50.6	DCR increased between 2005 & 2011
Independent Variables
Village Level Shocks
Drought	47.8	54.2	Village reported drought from 2006–2010
Flooding	18.8	33.6	Village reported flood from 2006–10
Hailstorm	20.5	17.8	Village reported hailstorm from 2006–10
Cyclone	8.4	17.6	Village reported cyclone from 2006–10
Household & Village Level Controls
Farmer household	39.7	35.4	Household’s primary occupation is farming
Land Class
Landless	39.2	39.3	0 acres land owned
Marginal or small farmers	44.2	54.4	0–5 acres land owned
Medium or large farmers	16.5	6.2	> 5 acres land owned
Irrigated land	32.4	41.3	Household owns irrigated land
Below-poverty ration card	45.0	38.9	Household has a Below Poverty Line Ration Card
Caste
Brahmin & forward castes	17.3	13.8	Brahmins and other upper castes
Other backward castes	46.9	42.6	Includes former Shudras or those performing menial tasks
Scheduled castes	26.9	24.5	Includes Dalits (formerly ‘Untouchables’)
Scheduled tribes	6.3	8.1	Indigenous communities
Others	2.7	11.1	Religious minorities
Woman-headed HH	3.2	3.6	Household is headed by a woman
Health centre > 5 kms	45.2	28.9	Nearest primary health centre > 5 kms from village
Hospital > 10 kms	54.6	54.8	Nearest government hospital > 10 kms from village
Bank > 5 kms	33.9	22.3	Nearest bank > 5 kms from village
Households (N)	10,426	11,059	HHs who have borrowed at least once in 2005/2011

Outcomes are measured as change from 2004–2005 to 2011–12, and control variables are measured in 2004–05. Split-off households were assigned the values of their natal household in 2004–05. At the village level, we use data on self-reported shocks such as droughts, floods, cyclones, and hailstorms, and measure exposure as the (binary) exposure to each shock from 2006 to 2010. These shocks were reported in the 2011 survey and serve as an alternative climate exposure in our analysis, but with a slightly smaller sample size than the main analysis due to missing data for some villages. The most common shock was drought (experienced by half of communities during the study interval), followed by hailstorms, flooding and cyclones. We also extract data on distance from key infrastructural facilities such as primary and community health centres, government hospitals, and banks as control variables.

Our primary information on climate exposures comes from NASA’s Modern-Era Retrospective Analysis for Research and Applications Version 2 (MERRA-2), which integrates a variety of station-based and remotely-sensed climate data products to produce a physically consistent model of the earth system with a spatial resolution of ½° latitude by ⅝° longitude (Galero, et.al, 2017). These data have previously been successfully used in a variety of applications examining the social consequences of climate (Gray and Wise, 2016). We extracted daily MERRA-2 data from 1981–2010 on precipitation rate and mean temperature for the spatial centroid of each IHDS sample district, the smallest spatial scale at which location information is available. We then use these data to define season-specific, five-year climate anomalies. First, we create multi-annual seasonal averages by measuring the mean precipitation rate and temperature from 2006–2010 for each of three seasons: kharif or monsoon (May-September), rabi or winter (October-January), and zaid or summer (February-April). We then standardize these values relative to all other five-year, seasonal averages in the dataset to create standardized climate anomalies. These values thus measure the extent to which our study period differed from the historical climate in each district and in each season. The mean district-level climate experienced during the study interval was primarily hot and wet relative to the historical climate, with the exception of monsoon temperatures which were relatively cool (Table 2).

Table 2.

Summary statistics for climate and other district-level variables.

	Arid Districts				Humid Districts
	Mean	SD	Min	Max	Mean	SD	Min	Max
District-level climate anomalies
5-yr summer temperature anomaly	0.50	1.45	−2.65	2.57	1.10	1.18	−4.34	2.64
5-yr monsoon temperature anomaly	−0.99	0.50	−1.94	1.71	−0.60	0.87	−1.70	2.00
5-yr winter temperature anomaly	0.74	1.45	−2.11	2.44	1.64	1.17	−1.70	2.84
5-yr summer rainfall anomaly	1.30	0.53	−0.44	2.17	0.73	0.66	−1.72	2.08
5-yr monsoon rainfall anomaly	1.81	0.38	−0.30	2.66	1.08	0.82	−1.89	2.44
5-yr winter temperature anomaly	1.75	0.62	0.44	3.04	0.77	0.87	−1.51	2.76
District-level controls
Historical rainfall 1981–2003 (mm/day)	2.87	1.20	0.67	8.63	3.49	1.38	0.92	9.56
Historical temperature 1981–2003 (Celsius)	26.73	1.14	22.49	28.87	24.90	3.66	−3.69	28.82
Area under cereals (%)	41.66	20.30	3.01	88.76	62.95	23.17	1.04	98.54
Area under pulses (%)	10.98	9.79	0.12	44.44	8.05	8.34	0.01	46.58
Area under sugarcane (%)	2.57	4.62	0.00	27.81	2.55	7.06	0.00	48.31
Area under fruits, veg, & spices (%)	5.57	4.66	0.22	23.21	8.18	7.33	0.00	50.35
Area under oilseeds (%)	11.68	10.03	0.00	44.90	5.94	6.87	0.00	33.31
Area under fibre crops (%)	5.34	7.81	0.00	31.66	1.27	2.96	0.00	11.74
Area under other crops (%)	21.85	11.67	0.00	48.88	10.98	12.76	0.12	62.56
Cropping intensity (%)	77.59	13.57	31.07	100.00	72.08	15.31	39.56	100.00
Districts	120				155

Relative to raw climate values, these standardized anomalies have the advantage of being uncorrelated (on average) with the historical climate and can thus be interpreted as natural experiments (Nordkvelle, J. et. al, 2017). The spatial pattern of the winter temperature anomalies is displayed in Figure 1. To account for any residual correlations between these climate exposures and local conditions, we also measure the mean temperature and precipitation rate from 1981–2003 (prior to the IHDS surveys) to include as controls.

Figure 1.

Spatial distribution of the sample districts, agroecological zones, temperature anomalies and household indebtedness

To further account for local agroecological conditions, we extract district-level data from the Indian Ministry of Agriculture on area under different crops, the net sown area (NSA), and the gross cropped area (GCA) for the year 2005. These data are collected by the Ministry via its field offices and then made available through a dashboard (MoA, 2020). From these data, we calculate the cropping intensity, which is the ratio of GCA to NSA expressed as a percentage, as well as the percent of cropped area in the following categories: cereals, pulses, sugarcane, oilseeds, fruits, vegetables, spices, and condiments, cotton, and other crops.

Finally, using the classification provided by the Indian Council for Agricultural Research (1996), we grouped the IHDS districts into two broad agroecological regions (AER): ‘Arid’, comprising hot arid and semi-arid AERs and encompassing most of peninsular, western and North-western India, and ‘Humid’, consisting of sub-humid, perhumid, and coastal regions, and comprising districts on eastern and western coasts, Gangetic plains, the hill states of Jammu and Kashmir, Himachal Pradesh, and Uttarakhand, and the states of eastern and North-Eastern India (Figure 1). We use this distinction to stratify the analysis and thereby observe whether and how climate-debt relationships vary across agroecological conditions. In doing so, we account for the distinct soil composition, landcover, and land-use that characterize each of these regions.

Measures of indebtedness

There are two aspects of indebtedness that are commonly quantified in the literature: size and depth (NSSO, 2014; Schicks, 2014; Chichaibelu, et.al, 2017). The first captures the total debt burden of a household, with the number of loans and the total monetary value of loans as the most common indicators. The limitation of using these measures is that they tell us what the debt burden of a household is, but not what the intensity of that burden is (Darling, 1928; Van Gunten and Navot, 2018). The depth or intensity of indebtedness signifies the total debt burden in relation to household resources such as assets or income and can be measured using financial ratios (Harness, et.al, 2008; Van Gunten and Navot, 2018). Commonly used financial ratios include debt-to-asset ratio (DAR), debt-to-income ratio (DIR), debt-service-coverage ratio (DSCR), and debt-to-consumption ratio (DCR) (Bhaduri, 1983; Schicks, 2013 & 2014; Chichaibelu, et.al, 2017). Making use of the measures available in IHDS, we examine DIR and DCR. DIR is the ratio of total outstanding debt to the gross annual household income, expressed as a percentage. In financial management, a DIR that is equal to or below 36% is considered ideal, whereas a ratio in excess of 36% indicates that any hit to income would leave a household struggling to repay its outstanding loans (Ross, 2019). DCR is the ratio of total outstanding loans to total annual consumption expenditure, expressed as a percentage. Although not widely used in financial management, DCR has been used in agrarian studies to indicate the extent to which a household’s consumption expenditure is fuelled by debt. A higher DCR suggests that a household is funding a larger share of the consumption expenditure through borrowings (Bhaduri, 1983). Due to lack of data on assets and debt payments in IHDS, we do not employ other measures such as the debt-to-asset ratio or the debt-service-coverage ratio.

To examine changes in household indebtedness over the study period, we define four dichotomous outcomes which measure whether the number of loans, the value of loans, the DIR and the DCR respectively increased over time, versus decreased or remained the same. Here, a dichotomous measure of change is preferred over taking the linear difference between these values over time because their distributions are highly skewed. Results of linear regressions of log-transformed continuous outcome variables (shown in Appendix Tables 6 & 7) broadly corroborate the findings from logistic regressions. All monetary values were adjusted for inflation between 2005 and 2011, using measures sourced from the website inflationtool.com (see Appendix 15 for details). The district-level DIR values in 2005 and 2011 are displayed in Figure 1, documenting a substantial increase in many districts.

Analysis

We use logistic regression to examine the effects of climate exposures on changes in indebtedness while accounting for potential confounders at the household, village, and district scales. The primary specification includes the district-level climate anomalies, and a secondary specification includes the village-level shocks. In both specifications, household-level controls include whether the principal occupation is farming, the land area owned, whether the household owns any irrigated land, whether the household has a below-poverty-line ration card, gender of the household head, and caste (Table 1). Parsimonious models for both specifications have been presented in Appendix tables 16 and 17. Village-level controls include whether there is a primary health center within 5 km, whether there is a government hospital within 10 km, and whether there is a bank within 5 km, capturing accessibility to key services. District-level controls include the historical rainfall and temperature (1981–2003), the percentage of cropped area under various crops, and cropping intensity (Table 2). To account for small numbers of missing values on the control variables, we also include variable-specific indicators for missingness in the regression, and in the case of missing values on continuous variables also interpolate the district-level controls to their state mean. The number and percentage of missing values have been furnished in Appendix Table 12. We also estimated our models without the missingness variables and did not find any significant difference in the direction or magnitude of our effects (for more, please refer to Appendix Tables 13 and 14). All analyses are stratified by agroecological zone, include sampling weights produced by IHDS, and account for clustered standard errors at the district level. To test for potential differential vulnerability to the climate effects across household characteristics, we subsequently allow the effects of climate anomalies to vary with indicators for farm household, land ownership, caste, and gender of the head by adding one set of interactions at a time to the primary specification.

Results

Table 3 presents the results of the primary specification examining the effects of climate anomalies on indebtedness across four outcomes and two agroecological zones. At the foot of the table, the test for joint significance reveals that the climate effects are jointly significant for all outcomes in both zones. Among the climate effects, we find most notably a strong association between temperature anomalies during the winter season and household indebtedness, and that the direction of these effects differs across agroecological zones. A one-unit increase in winter temperature anomalies in the arid sample (equivalent to a one standard deviation increase over the historical value) results in a 33% increase in the odds that the number of loans borrowed will go up (p = 0.007), a 25% increase in the odds of the loan amount going up (p = 0.006), a 23% increase in the odds of DIR going up (p = 0.030), and a 22% increase in the odds that DCR will go up (p = 0.020). In contrast, in the humid sample a one-unit increase in winter temperature anomalies results in a 14% decrease in the odds of loan amount increasing (p = 0.093), a 17% decline in the odds that DIR will go up (p = 0.083), and an 20% decline in the odds that DCR will increase (p = 0.030). We discuss these contrasting results below.

Table 3.

The effects of climate anomalies on indebtedness in arid and humid regions.

	Arid Districts				Humid Districts
	Increase in loans	Increase in debt	Increase in DIR	Increase in DCR	Increase in loans	Increase in debt	Increase in DIR	Increase in DCR
Climate anomalies
Summer temperature	0.80*	0.75**	0.72**	0.75**	1.00	0.94	1.01	0.99
Monsoon temperature	1.50*	1.34	1.76*	1.71+	1.08	1.15	1.02	0.99
Winter temperature	1.33**	1.25**	1.23*	1.22*	0.83	0.86+	0.82+	0.80*
Summer rainfall	0.82	0.99	0.85	0.92	1.19	1.22*	1.33**	1.42***
Monsoon rainfall	0.57*	0.95	1.00	1.20	0.90	0.87	0.84	0.71+
Winter rainfall	1.48*	1.25	1.29	1.13	1.12	1.15	1.19	1.11
Household and village controls
Farm household	0.93	0.91	0.86	0.93	1.08	1.20+	1.23*	1.34**
Landless	1.35*	1.24	1.12	1.23	0.82	1.14	0.94	1.13
Marginal or small landholders	1.07	1.29*	1.07	1.22+	0.78+	1.03	0.88	0.99
Irrigated land	1.01	0.93	1.00	0.93	0.72**	0.84*	0.75**	0.75**
Ration card	0.81	1.05	0.93	0.98	0.66***	0.75***	0.71***	0.75***
Other backward caste	0.99	0.95	0.96	0.95	0.93	1.00	0.88	0.93
Scheduled caste	0.99	0.92	0.89	0.93	0.91	1.02	0.94	0.97
Scheduled tribe	1.10	0.89	0.94	0.90	0.79	0.64**	0.75+	0.77+
Minorities & others	0.76	1.15	1.17	1.14	0.76	0.88	0.82	0.84
Woman-headed HH	0.96	1.19	0.72+	1.05	1.10	1.15	0.92	1.06
Health centre > 5 km	1.24+	1.14	1.27*	1.17	0.82+	1.02	0.97	0.91
Hospital > 10 km	0.79+	1.07	0.91	0.97	0.68**	0.79*	0.78*	0.78*
Bank > 5 km	0.90	0.94	0.88	0.95	1.05	0.82*	0.80**	0.82*
District controls
Historical rainfall (mm/day)	0.56***	0.71*	0.67**	0.68**	1.28*	1.16+	1.22*	1.12
Historical temperature (Cel)	0.92	1.02	1.05	1.02	0.99	1.00	1.00	0.98
Area in pulses	1.00	1.00	1.00	1.00	0.97**	0.98+	0.98*	0.97*
Area in sugarcane	0.96**	0.99	1.00	1.00	0.97**	0.99*	0.98**	0.97***
Area in fruits, veg, & spices	0.99	0.96+	0.95+	0.95*	0.94**	0.98*	0.98*	0.98*
Area in oilseeds	1.02	1.00	1.00	1.00	1.00	1.00	1.00	1.01
Area in fibre crops	1.01	0.99	1.00	1.00	1.04	1.05+	1.05	1.07*
Area in other crops	0.99	1.01	1.00	1.00	0.99	0.98**	0.98*	0.98**
Cropping intensity	0.98**	0.99**	0.98***	0.98**	1.00	1.00	1.00	1.00
Joint test of anomalies	12.14+	19.8***	21.32***	34.33***	10.51	33.29***	30.52***	31.85***
Observations	10,426	10,426	9,594	9,329	11,059	11,059	9,929	9,481

^***p<0.001

^**p<0.01

^*p<0.05

⁺p<0.10

Constant and missing indicators included but not shown.

We also find weaker and zone-specific associations between the other climate anomalies and changes in household indebtedness. Temperature anomalies during the monsoon and summer are particularly important in arid regions. In arid areas, but not in humid areas, a one-unit increase in monsoon temperature anomalies leads to a 50% increase in the odds of loan number increasing (p = 0.031), a 76% increase in the odds of DIR going up (p = 0.048), and 71% increase in the odds of DCR going up (p = 0.072). Notably acting in the opposite direction, a one-unit increase in summer temperature anomalies in the arid sample results in an 20% decline in the odds of an increase in loan number (p = 0.042), a 25% decrease in the odds of an increase in the loan amount (p = 0.002), a 28% decline in the odds of higher DIR (p = 0.002), and a 25% decline in the odds of a higher DCR (p = 0.003). The association between rainfall and household indebtedness also shows mixed results. In the humid sample, a one-unit increase in summer rainfall anomalies results in a 22% increase in the odds of the loan amount going up (p = 0.045), a 33% increase in the odds of DIR going up (p = 0.004), and a 42% increase in those of DCR going up (p = 0.001). In the arid sample, a one-unit increase in the winter rainfall anomaly leads to a 48% increase in the odds of the number of loans going up (p = 0.039) whereas a one-unit increase the monsoon rainfall anomaly leads to a 43% decline in the odds of the loan number going up (p = 0.045).

Table 3 also presents the results for the control variables, revealing that the association between household characteristics and indebtedness varies between arid and humid samples. In the humid sample but not the arid sample, farm households were more likely to experience increasing loan numbers, loan amounts, DIR and DCR relative to non-farm households. For the arid sample but not the humid sample, landless households more often saw increases in the number of loans and marginal/small landowners more often experienced increases in the loan amount and in DCR. In the humid sample, households that own irrigated land were less likely to experience increasing indebtedness as measured by the number of loans, the DIR and the DCR. The possession of a below-poverty-line ration card, which enables access to subsidized food grains and other essential household commodities, was associated with a decline in all four parameters of indebtedness in humid districts. The effects of caste, gender of the head, and accessibility are relatively weak and not consistently significant across the four outcomes. For robustness, we examined the effect of different levels of climate anomalies on household indebtedness. We found that households with > +1 standard deviation increase in winter time temperature reported a significant increase in their debt burdens (for more refer to Appendix tables 19 and 20).

The effects of the district-level controls also vary across agroecological zones. In arid regions, increasing indebtedness was less common in districts with higher historical rainfall and with higher cropping intensity, suggesting that agriculturally marginal areas fared worst. In humid regions, indebtedness was more likely to increase in wetter districts and in those with a lower fraction of cropped area in cereals (the reference category), suggesting benefits from moderate levels of rainfall and from agricultural diversification.

In Table 4 we present the results for the second specification in which we examine the impact of village-level climatic shocks such as droughts, floods, hailstorms, and cyclones on household indebtedness. Households from villages in arid regions that reported a drought during 2006–10 saw a 47% increase in the odds of a higher loan amount (p = 0.001), a 51% increase in the odds of a higher DIR (p = 0.001), and 43% increase in the odds of a higher DCR (p = 0.004). In contrast, we find that the occurrence of hailstorms in arid districts is associated with a decline in all four measures of household indebtedness, which potentially reflects their association with other climatic conditions such as temperature. The other effects of extreme events are largely statistically non-significant or inconsistent across outcomes. We ran pair-wise correlations to check if village-reported shocks coincided with anomalies in temperature and rainfall. There is a small positive correlation between summer temperature anomalies and occurrence of droughts in both arid/semi-arid and humid/sub-humid regions. There is also a small correlation between increase in summer temperature anomalies and village-reported occurrence of hailstorms (for further details, refer to Appendix Table 5).

Table 4.

The effects of village-level shocks on indebtedness in arid and humid regions.

	Arid Districts				Humid Districts
	Increase in loans	Increase in debt	Increase in DIR	Increase in DCR	Increase in loans	Increase in debt	Increase in DIR	Increase in DCR
Drought	1.00	1.47***	1.51**	1.43**	0.78*	1.01	0.98	1.00
Flood	0.98	0.89	0.88	0.96	0.86	0.92	0.91	0.94
Hailstorm	0.76*	0.77*	0.71**	0.75*	1.14	0.97	1.00	1.01
Cyclone	1.00	1.13	1.13	0.98	0.80	0.97	0.89	0.95
Joint test of shocks	7.73	20.95***	17.13***	12.6*	9.27+	0.92	1.86	0.53
Observations	9,767	9,767	9,314	9,054	9,516	9,516	9,355	8,964

^***p<0.001

^**p<0.01

^*p<0.05

⁺p<0.10

All controls, constant and missing indicators included but not shown.

Appendix Table 1 presents the net climate effects for each household category from the analysis of potential differential vulnerability across household characteristics. Joint tests of significance of the interactions reveal that all four characteristics significantly modify the climate effects in both arid and humid regions, but the largest differences are across land ownership and caste in arid regions and thus we focus on those here. In arid/semi-arid regions, the odds of higher loan number, loan value, DIR and DCR for landless households increase by 34%, 28%, 26%, and 20% respectively with each unit of winter temperature, versus 30%, 16%, 17% and 18% for marginal and small farmers. Similarly, these odds increase by 38%, 24%, 20% and 22% for Other Backward Caste (OBC) households and 72%, 44%, 44%, and 50% respectively, for Scheduled Tribes (ST), versus 39%, 29%, 22% and 21% for upper caste households. Scheduled Caste (SC) households experience the largest increases in indebtedness with monsoon temperatures and the largest decreases in indebtedness with summer temperatures.

Across land ownership categories, the effects of monsoon temperatures are largest for marginal and small landowners and for SC, ST, and OBC households. In short, land-poor and lower-caste households appear to be more vulnerable to increasing indebtedness in arid regions due to climate anomalies.

To test the robustness of the logistic regressions, we analysed the raw differences using ordinary least square regression. For each of the outcome variables, ‘loan amount increase’, ‘DIR increase’, and ‘DCR increase’, we computed the raw difference and then logged the values, to account for their skewed distribution; before performing OLS regression (see Appendices 5 and 6 for further details). These results are broadly consistent with our main findings. We also examined whether climate anomalies push households towards overindebtedness, defined here as a state where the DIR and DCR are above 36% respectively. We found a significant association between increase in monsoon temperatures and summer rainfall anomalies in humid/sub-humid regions and the odds of DIR and DCR going over the overindebtedness thresholds. In arid/semi-arid regions, we found that households in villages that reported droughts were more likely to see their DIR and DCR breach the overindebtedness thresholds (for more on this, see Appendices 7 and 8). Lastly, we tested whether the interaction of climate anomalies and district level cropping patterns resulted in an increase in household indebtedness. We found that an increase in cropping intensity when combined with an increase in monsoon and winter temperature in arid/semi-arid regions increased household debt burden (refer to Appendix Table 2).

Discussion

The results reveal that climate exposures have substantial and multidimensional consequences for household indebtedness, particularly in semi-arid areas. In these areas, winter temperatures increase all aspects of indebtedness, and monsoon temperatures similarly increase three out of four measures of debt, effects which are particularly strong for landless and low caste households. Summer temperatures however have the opposite effect, a contrast we return to below. The effects of rainfall are weak and inconsistent. In humid areas, there is a distinct pattern in which winter temperatures reduce all measures of indebtedness whereas summer rainfall largely has the opposite effect. What are the mechanisms that drive these associations? Why do increases in temperature increase the size of debt for rural households in the drier parts of India? Below, we present possible pathways through which climatic shifts could be affecting household debt in the two regions.

Climate-agriculture-indebtedness pathway

One likely pathway through which climate change could trigger shifts in household indebtedness is via crop productivity. An increase in temperature is known to depress crop yields, as is untimely rainfall. In arid/semi-arid regions, even a 1°C increase in temperature can result in a significant decline in crop yields (Palanisami, et.al, 2019). Similarly in humid/sub-humid regions where wheat is a major crop, the slightest increase in temperature can reduce the yields of this heat-sensitive crop (Palanisami, et.al, 2019). Wheat is typically harvested in early summer (Sarthi, 2019; Agropedia, 2020) and rainfall at harvest time can lead to significant crop losses. Further, untimely rainfall can engineer crop losses by triggering flash-floods, a common phenomenon in the mountainous districts of Himachal Pradesh and Uttarakhand in Northern India and in districts from North-Eastern states like Assam (DTE, 2020; Sarkar, 2020). Crop losses translate into income losses for those households that are dependent on farming and allied activities. In such a scenario, debt can be an effective tool to overcome income disruptions and maintain household consumption levels.

Another way climatic shifts can increase household indebtedness is by inducing households to make investments in adaptive capacities. For instance, farmers are likely to invest in pumps, drip-irrigation kits and sprinkler systems to counter the loss of soil-moisture resulting from high temperatures or recurring droughts. Along with irrigation, farmers are also likely to increase investments in fertilizers and improved seeds to increase yields in the face of mounting climate-related uncertainties (Birthal, et.al, 2015;Taraz, 2018). Through these investments, farmers try to infuse greater predictability into agriculture, and they borrow in order to do so. In our data, we found that 54–55% of the households that experienced increased their area under irrigation also saw an increase in all parameters of indebtedness (for more, see Appendix Table 11).

This pathway could partly explain why indebtedness tends to increase with winter and monsoon temperatures and village-level droughts in semi-arid areas, as well as high rainfall in humid areas. A decline in yields of important crops like rice and wheat, which comprise a major percentage of the total area under cultivation, would translate into a loss of income for farming households and for households that depend on agricultural wage labour.

Climate-health-indebtedness pathway

Healthcare expenditures are another pathway through which climatic shifts can impact rural household indebtedness. Increases in temperature have been associated with a rise in the incidence of heat-related illnesses and cardiovascular stress (Crowley, 2016; Lundgren-Kownacki, Karin, 2018; Sahu, et.al, 2013). An increase in monsoon temperature can result in higher incidence of waterborne diseases like cholera, typhoid, and diarrhoea; and vector-borne diseases like malaria and chikungunya (Dhara, et.al, 2013; Sahu, et.al, 2013; Crowley, 2016; Lundgren-Kownacki, Karin, 2018; Narayan, et.al, 2020). Rural households in arid/semi-arid regions also tend to have a higher dependence on manual labour, whether as farmers or as daily-wage labourers, and are therefore more susceptible to heat-related illnesses. High healthcare costs in turn have been shown to be a major drain on the finances of rural households, driving them deeper into debt and impoverishment (Krishna, 2017), and, in extreme cases, suicide (Sethi, 2018). Scholars have further demonstrated that unforeseen medical expenses are a major reason why rural households borrow (Ramachandran and Swaminathan, 2002; Pradhan, 2013). We did not find significant evidence of medical expenses increasing household debt. However, a majority of those who saw an increase in their medical expenses also experienced an increase in their overall level of indebtedness (see Appendix Tables 10 and 11).

But what explains the negative relationship between winter temperature and household debt in humid/sub-humid temperature? In humid regions, a fall in winter temperature has been found to result in an increase in the incidence of respiratory illnesses (Narayan, et.al, 2020). The corollary is that a warmer winter resulting from an increase in temperature would translate into a reduction in the incidence of such illness. This could explain why rural household borrowing goes down with a corresponding increase in winter temperatures.

Climate-vulnerability-indebtedness pathway

We also find that climate anomalies have differential impact on the indebtedness of households belonging to different socioeconomic groups. In arid regions, the influence of winter temperature and rainfall anomalies on debt is stronger for ST and OBC households relative to upper-caste households, and this also true for the influence of summer and winter temperatures on landless and small-holding households relative to large landowners. A closer look at the IHDS data offers some explanations for these trends. A significant percentage of ST households—about 45%—depend on agriculture for their livelihoods (Appendix Table 3). At the same time, fewer households from this group have access to irrigation (23%; Appendix Table 4). Irrigation helps farmers cope with fluctuations in rainfall and temperature. A significant proportion of ST households (45%) are also dependent on wage labour—agricultural and non-agricultural—for their livelihoods. Higher temperatures can expose people involved in physical labour to a host of heat-related health hazards, including sunstrokes and dehydration (Lee and Lim, 2018; Lundgren-Kownacki, Karin, 2018). As discussed above, such hazards are likely to force households to lose workdays and potentially rely on debt to overcome the disruptions to their cashflows.

Households belonging to marginalized caste groups (ST and OBC) and those who were landless or small landholders are more likely to rely on moneylenders, large landholders, seed and agricultural commodity traders, and government agencies to access the means of production needed to sustain household reproduction (Bhaduri, 1986). Shifts in rainfall and temperature are likely to push small farmers to make significant investments in adaptive capacities. These could include improved seeds, irrigation, or more fertilizers. All of this requires capital which these households do not have access to, and hence must borrow. Uncertainties in agriculture, rooted in climate change, would translate into a decline in the availability of work for rural wage workers, forcing them to look outside their villages for wage labour, possibly through middlemen. Working as wageworkers, often under highly exploitative terms, often causes households to borrow to meet consumption needs and thus getting caught in debt traps (Natarajan, 2018; Carswell, 2020). In this way too, climate change can increase indebtedness for the rural poor.

In summary, it appears likely that the processes we observe results from multiple, interacting pathways from climate to debt, which together contribute to the hazardscapes (Ribot, 2014) of rural households, particularly for marginalized groups.

Conclusion

By bringing together the literatures on agrarian change and climate vulnerability in India, and by using a large-scale, and longitudinal approach, we show that climate anomalies and village-level climate shocks both contribute to increasing indebtedness in rural India. Notably, we identify winter and monsoon temperature anomalies as well as village-level droughts as contributing to growing household debt in arid and semi-arid areas. These findings indicate that ongoing shifts in rainfall and temperature under climate change are likely to increase the debt burden on rural households, particularly those in dry areas and belonging to marginalized groups.

Debt is a vital instrument for coping with income losses, achieving short-term consumption smoothing, meeting emergency expenditure, and making investments in agriculture and allied activities. It is also important for building the adaptive capacities of rural households so that they can better cope with shifts in climatic patterns and the resultant disruptions. However, debts taken beyond the repayment capacities of households can also trap them into a cycle of poverty, a cycle that is likely exacerbated by climate change. Thus, policies that seek to address the problem of rural indebtedness must consider the role of climate change; and those seeking to tackle the consequences of climatic shifts must spare a thought to the important, yet sometimes corrosive, role that debt can play in the wellbeing of rural households.

This work opens multiple avenues for future research that lie at the intersection of rural sociology, agrarian studies, and climate vulnerability. How do debt and climate change co-evolve over longer periods of time? Do past climate shocks or existing debt make households vulnerable to additional climate change? How does climate influence other agrarian and economic outcomes such as depeasantization, landlessness, rural household assets, and financial inclusion? How can anti-poverty and financial inclusion policies be crafted so that they can help buffer the rural households against the deleterious effects of debt and climate change? Given the increasing availability of large-sample and longitudinal data on these topics in low-resource settings, there is much more that we can learn.