Financial inclusion as a pathway to welfare enhancement and income equality: Micro-level evidence from Nigeria

ABSTRACT

While the importance of financial inclusion as a means of poverty and income inequality reduction has long been recognised, the paths to welfare enhancement and income equality through financial inclusion remain partially acknowledged. Using micro-level data on 1 750 rural Nigerian households, this study examines the finance-welfare nexus by constructing a multi-variable financial inclusion index. The results first show that financial inclusion exerts a strong positive influence on household welfare. However, the decomposition analysis shows that middle- and high-income households gain more from financial inclusion in comparison to the targeted low-income households. Second, informal livelihood strategies, such as environmental resource extraction, crops, and livestock production, revealed reduced welfare disparities across income distributions. Therefore, for financial inclusion to alleviate welfare inequality and ensure income convergence, rural financial markets must be redesigned to allow wider access to credit, specifically for low-income and vulnerable households.

KEYWORDS:

• Financial inclusion
• welfare
• inequality
• economic activities

JEL Codes:

• D3
• H5
• I3
• P16

1. Introduction

Financial inclusion is one of the social concepts that is controversially defined among different disciplines. The economic conception of the term was orientated within the broader context of inclusive development that considers inclusivity in financial services as an important means to tackle poverty and inequality (Chibba, 2009). Financial inclusion is thus defined as the access to useful and affordable financial products and services that meet the financial needs of low-income and vulnerable members of society (World Bank, 2018). This is based on the premise that financially included individuals are more favourably disposed to invest in education, start and expand businesses, manage risks and absorb financial shocks than financially excluded individuals (Banerjee et al., 2015; Krumer-Nevo et al., 2017).

On the other hand, geographers are mainly concerned with the physical access to banking services driven as a result of either the availability of formal financial service providers or limited access due to the closure of bank branches (Leyshon & Thrift, 1995). While the availability of banking services would stimulate access to banking, closure of banking branches would cause some individuals or certain groups of people to discontinue accessing the financial service infrastructure (Wentzel et al., 2016). Moreover, sociologists strongly believe that financial exclusion is as an important contributor to vulnerability that could possibly lead to social inclusion. Sinclair (2013) argued that access to financial services is essential for citizens to be economically and socially integrated into today’s society. The implication is that individuals can improve their welfare through the increased financial access, which could have spill-over effects on the overall prosperity of their communities and the economy at large (Ibrahim, 2014).

Financial exclusion as a polar opposite of financial inclusion is characterised by the inability of individuals to access essential financial services that meet their financial needs (Sinclair, 2013). Conroy (2005) contended that financial exclusion (or deprivation) is a process that prevents poor and disadvantaged social groups from gaining access to the formal financial system. While access to transaction accounts is considered as the most basic form of formal financial inclusion, being financially excluded means that transactions by individuals, households and enterprises are entirely conducted in cash, and this could increase their susceptibility to irregular cash flows (Wentzel et al., 2016).

Indeed, much of the debate and controversies over financial inclusion are built around the impact of micro-credit on the reduction of poverty and welfare disparities. This argument is based on the promising performance of financial services on welfare augmentation, particularly in a liberalised financial system. Formal participation in a financial system is suggested to offer households immunity against idiosyncratic risks and sudden shocks, because with access to cheap finance, poor households would be able to invest in education and diversify their livelihoods (Banerjee et al., 2015; Giordano & Ruiters, 2016).

With almost 350 million out of the estimated 2.5 billion unbanked adults dwelling in Sub-Saharan Africa (SSA) (Pagura & Kirsten, 2006; Enhancing Financial Innovation & Access [EFInA], 2017; World Bank, 2018), Africa is one of the continents at the risk of not meeting the goal of reducing financial exclusion to 20% by the year 2020. Among the major African economies, South Africa has the highest proportion of banked and the lowest proportion of financially excluded adults, with 77% and 13% in 2015, respectively; however, although Nigeria has a relatively large banked population (38.3%), it also has the highest proportion of financially excluded adults (41.6%) (EFInA, 2017). Given that Nigeria is particularly well-positioned (in terms of human, financial and natural resources) to reap the benefits associated with advancing financial accessibility among the poor and disadvantaged groups, the country provides an important case study to examine the determinants of financial inclusion as well as to explore the pathways through which financial inclusion is enhancing welfare and reducing the income inequality of households that engage in various livelihood strategies.

The remainder of this paper is organised as follows: the subsequent sections of this introduction contain an overview of the Nigerian financial system and financial inclusions reforms, as well as the body of relevant literature. Section two deals with the methodological issues consisting of a detailed econometric model and empirical variables. Section three describes the data collection and presents the descriptive statistics. Thereafter, the results and discussion are presented in Section 4, and the final section provides a conclusion and policy considerations.

1.1. The Nigerian financial system, financial inclusion reforms and implication on welfare

Nigeria’s financial system consists of the formal and informal sub-sectors, with the central bank assuming the regulatory and supervisory roles on the 26 players in the formal segment (i.e. 21 commercial banks, four merchant banks and one interest-free bank) (Central Bank of Nigeria, henceforth CBN, 2017). On the other hand, the informal sector that is dominated by the local moneylenders, self-help groups, and rotating savings and credit associations (ROSCAs) substantially covers a wide range of market activities within the traditional sector in which earning opportunities are limited (CBN, 2012; Ibrahim, 2014). While the formal financial market provides services to about 38% of the economically active population, the remaining 62% excluded populace are either accessing finance from the informal financial sector or have no current access to finance at all (CBN, 2017).

Financial inclusion policy can be traced back to the pre-structural adjustment programme (SAP) rural banking programme of the 1970s where all the then 18 commercial banks were directed to open rural branches (CBN, 2012; Dimova & Adebowale, 2017). This was based on the premise that having more rural bank branches would inculcate the desired banking habit needed to enhance the welfare of the rural populace. In this connection, new commercial banks were licenced, which increased their number from 18 in 1977 to 41 in 1985, with the number of rural branches rising from 200 to 300. Given the number of branches established under the rural banking scheme, there is no doubt that the programme helped to promote the banking habit in rural areas. The question of whether the scheme achieved the policy objective is however more contentious (Uche, 1999; Ibrahim, 2012). Some of the challenges that constrain the operation of the rural banking scheme lie in the lack of commitment of the CBN to particularly liaise with the local government authorities, to improve the credit attractiveness of the rural populace other than directing conventional banks to open more branches in rural areas (Uche, 1999; Ibrahim, 2014).

The economic policy package (SAP) which introduced a deregulation policy helped to sustain the phenomenal rise in the number of Nigeria’s commercial banks, which resultantly tripled the number of banks in six years (Dimova & Adebowale, 2017). The aim of the reform was to liberalise the financial system in order to drive the competitive flow of finance to the various sectors of the economy (Aliero & Ibrahim, 2012). At that point in time, the government realised the unsuitability of the conventional banks to effectively serve the financial needs of the rural poor. It thus encouraged the establishment of the unique financial institutions (such as community banks, people’s banks, etc.) charged with the responsibility of rural development through the provision of critical rural financial infrastructures. The failure of these rural financial intermediaries was partly attributed to corruption, inept management, inconsistent macroeconomic policies and political instability, among others (Uche, 1999; Dimova & Adebowale, 2017).

Following the banking crisis in the aftermath of the financial reforms in the late 1990s, there was a push for fine-tuning the reforms in line with international best practices (CBN, 2012; Ibrahim, 2014). Aside from measures such as banking consolidation as a result of the increase in capitalisation requirements and the formation of monitoring organisations like the Economic and Financial Crimes Commission (Dimova & Adebowale, 2017), the key highlight of the changed policy environment was the formulation of the Microfinance Policy, Regulatory and Supervisory Framework in 2005. This, as a reconstructive policy, recognised the need to subject the existing informal institutions to the CBN regulatory framework (CBN, 2012). The policy provided the stimulus for depth and the stability of the financial system. For instance, the stylised measure of financial inclusion indicates that the proportion of financially included adults increased from 23.6% in 2008 to 48.6% in 2014 (EFInA, 2017). This was partly achieved as a result of the technical guidance provided by World Bank, which led to the successful implementation of the National Financial Inclusion Strategy in 2012, as well as partly due to the sustainable increase in awareness and understanding of financial products and services through the National Financial Literacy Framework of 2012 (CBN, 2017). However, due to the increasing dependence on rain-fed cropping and pastoral cattle production led rural dwellers to heavily rely on microfinance and other informal finance infrastructure. Other sources of financial inaccessibility in rural areas include the low human capital and demographic challenges (Aliero & Ibrahim, 2012), infrastructural bottlenecks (e.g. road, market, internet, electricity, etc.), and the limited options in the interest-free financial market.

1.2. Financial inclusion, welfare and related empirical literature

The welfare of households to meet basic needs and sustain increases in income largely depends on access to formal financial services. There is rigorous empirical evidence that indicates a strong positive relationship between financial inclusion and income growth (Demirgüç-Kunt & Levine, 2008; Ibrahim 2014; Wentzel et al., 2016), and access to finance and poverty eradication (Beck et al., 2004). Interestingly, in the financial development literature, the findings of both micro- and macroeconomic empirical research are consistent. From the microeconomic viewpoint, studies have shown that financial inclusion positively affects welfare (Banerjee et al., 2015; Dimova & Adebowole, 2017). The literature on financial products even shows that access to formal financial services, such as demand deposit, micro-credit, payment facilities, and micro-insurance, increases household income, empowers women, smoothens consumption, and reduces vulnerability to financial shocks (Aliero & Ibrahim, 2012; Park & Mercado, 2015). Macroeconomic studies, on the other hand, have found a strong positive relationship between financial inclusion and overall human development (Chibba, 2009; Nuruddeen & Ibrahim 2014; Dimova & Adebowle, 2017). However, notwithstanding the growing body of literature, the channels through which formal financial services are affecting poor households continue to be inadequately understood (Demirgüç-Kunt & Levine, 2008; Banerjee et al., 2015).

The role of the financial sector as a leading contributor of growth has been widely accepted (Johnson & Nino-Zarazua, 2011) and over the last two decades, the focus has turned to solidifying the weak nexus between finance and poverty reduction, as well as the repositioning of the key players of the financial system by enhancing their capacity for deepening the financial infrastructure for better outreach. There is important literature on the effect of financial exclusion on the level of welfare, poverty and income inequality (Conroy, 2005; Kirsten, 2012; Sinclair, 2013; Wentzel et al., 2016; Krumer-Nevo et al., 2017). These studies have strongly emphasised the need to enhance the financial inclusivity of individuals, households and enterprises that could pave the way for the attainment of inclusive development. Other sets of studies (Uche, 1999; Baumann, 2004; Daniels, 2004; Pagura & Kirsten, 2006; Brannen & Sheehan-Connor, 2016) have examined how microfinance and other regulated non-banking institutions could complement the conventional banks in stimulating the banking culture, particularly for rural development.

Extant literature has already provided insights into the welfare implications of financial inclusion, particularly for African countries. However, the paths to welfare enhancement through financial inclusion remain partially acknowledged. A decomposition analysis of the impact of financial inclusion on livelihood activities, particularly within the context of a welfare drive, has policy implications for the understanding and determination of the best strategies for poverty alleviation that could enhance income and resilience in an increasingly shock-prone global economy. This paper adds to the literature by examining the pathways through which financial inclusion ensures income convergence and enhances welfare parity.

2. Empirical methodology

2.1. Econometric model

Assume a continuous welfare distribution as a function of the income increase as a result of financial inclusion, in the sense that each welfare effect of financial inclusion, as opposed to exclusion, can be represented by the financial status of a household that can sustain a decent welfare $[w,w+\mathrm{\Delta }w]$, where$\mathrm{\Delta }w\to 0$. Let $f\left(w\right)$ be the cumulative welfare of the household. Then, $i-f\left(w\right)$ is the ith household, whose income is greater than w. Hence, $i-f\left(w\right)$ represents a household with a decent living standard, subject to financial status$[w,w+\mathrm{\Delta }w]$. The argument is that financial deprivation (exclusion) is an increasing function of an income below w. This is based on the premise that poor and vulnerable households are often excluded from formal financial services due to moral hazard, information asymmetries, and the exorbitant costs of incorporating them into a formal financial system. A simple financial inclusion (FI) model can thus be established as:(1) $FI=\underset{lris}{\overset{hris}{\int }}z[i+f(w\left)\right]\delta x.$(1)

Note that $z[i+f(w\left)\right]$ is the financial inclusion for (or deprivation $z[i-f(w\left)\right]$ for not) having$[w,w+\mathrm{\Delta }w]$. If welfare is assumed to be an increasing function of per capita expenditure (or broadly well-being assets), the notation hris can be seen as returns from a high-income earning strategy sufficient for$f\left(w\right)$. If we observe that$w\left(F{I}_1\right)>w\left(F{D}_0\right)$, then it suffices to argue that financial inclusion is the potential source of the welfare differences between financially included households and financially excluded ones. The cumulative welfare of financially included household $w\left(F{D}_1\right)$is represented by a score of 1 as opposed to 0 for cumulative welfare $w\left(F{D}_0\right)$ for financially excluded household. The financial inclusion must exceed a cut-off of 50%, which is the equivalent of half of the weighted indicators used to distinguish between the two groups of households. The maximum inclusion score is 100%, given that each component of financial inclusion is equally weighted, as described in Table 1.

Table 1. Variables used for constructing financial inclusion.

Component (weight)	Indicators (weight)	Included if
Microcredit (1/4)	Agro-credit (1/8)	Household accesses agricultural loan
	Other credit (1/8)	Household benefits from other forms of empowerment loans
Demand deposit (1/4)	Savings account (1/8)	Household owns a saving account
	Current account (1/8)	Household uses transaction account
Insurance (1/4)	Micro-insurance (1/4)	Household is insured
Time deposit (1/4)	Fixed account (1/4)	Household invests in a time deposit account

Financial development literature was built on the ‘non-excludability assumption’, meaning there are no rules whatsoever barring a household from participating in the formal financial market, other than ignorance, financial illiteracy, irrationality, and other characteristics within the household domain. Given this assumption, the overall welfare distribution $\left(w\right)$ of financially included households $\left(F{I}_w\right)$ and that of financially excluded households $\left(F{D}_w\right)$ can be decomposed into explained/composition and the unexplained/welfare, as expressed in Equation (2):(2) $\underset{Overalleffect}{\underbrace{(F{I}_w-F{D}_w)}}+\underset{welfareeffect}{\underbrace{(F{I}_w-W_c)}}+\underset{compositioneffect}{\underbrace{W_c+F{I}_w}},$(2) where $W_c$ is the welfare counterfactual, which accounts for the welfare loss of financially-excluded households. To establish the paths to welfare augmentation of the various livelihood activities, we decompose the low-return income earning (lris) and high-return income earning (hris) into the proportion of earnings associated the various livelihood strategies and examine their impact on a household’s welfare. Hence, we analyse the system of equations:(3) $W_{1i}=\alpha F{I}_i+\beta Q_i+{\mu }_i,$(3) (4) $W_{2i}={\gamma }_0+{\gamma }_1{E}_i+{\gamma }_2{A}_i+{\epsilon }_i.$(4)

In Equation (3), $W_i$ is the level of welfare of the ith household (using two proxies – per capita expenditure and well-being index). These proxies were proved efficient in previous studies (e.g. see Dzanku, 2015, Gautam & Andersen, 2016); $F{I}_i$ measures financial inclusion (computed by factors explaining household’s access to financial services, as highlighted in Table 1); $Q_i$ is a vector of relevant control variables, which includes per capita income, household age (measure in year), gender, household size, and literacy (years of formal school); $\beta$ is the associated vector of the coefficients; ${\mu }_i$ is a white-noise error term, which is identically and independently distributed, with zero mean and constant variance. In Equation (4), $E_i$ denotes the vector of monetary earnings from various livelihoods activities, which consists of the profits from own businesses, off-farm wages, farm wages, earnings from crop production, livestock rearing, and environmental resource extraction; $A_i$is a categorical variable that controls for unobserved time-invariant characteristics, such as location where ith household is (1 = crisis prone village, 0 = otherwise); while ${\gamma }_i\text{and}{\epsilon }_i$ are the vector of estimated coefficients and a white-noise error term ${\epsilon }_i\sim N(0,\sigma )$, respectively.

There are two salient econometric drawbacks associated with the two pairs of preceding equations: simultaneity and endogeneity. Most households are engaging in multiple livelihood activities. This simultaneity of income streams could lead to the autocorrelation of the error term, such that$\text{Cov( }{\epsilon }_i,{\epsilon }_j|E_i,E_j)\ne 0$. On the other hand, endogeneity arises in the sense that financial inclusion could lead to more income earning opportunities and, thus, higher welfare for financially included households. However, it remains plausible that an income increase could potentially allow households greater access to finance. To eliminate these problems, we first used primary data, where a restriction is imposed on a single option in designing the question on primary occupations and its associated income in the questionnaire. Second, the potential endogeneity issue was addressed by applying the propensity score matching (PSM) of Rosenbaum and Rubin (1983), based on multiple algorithms, as suggested in Caliendo and Kopeinig (2008).

The ordinary least squares (OLS) method was applied in the estimation of the parameters from the preceding equations. Moreover, owing to the deficiency of OLS capturing variable inter-relationships at different points in the conditional income distribution, we further apply quantile regression. In testing robustness, the counterfactual decomposition proposed by Machado and Mata (2005) was applied. Households were decomposed into financially included and excluded. Then, two proportional multipliers were used to examine the estimated coefficients: the coefficient effect measures the extent to which differences in income across quantiles are driven by financial inclusion rather than other covariates, while the characteristics effect measures the extent to which household characteristics contribute to the various income differences.

2.2. Empirical variables

The empirical measurement of variable FI depends on it conceptual definition. Two issues are relevant to conceptualising financial inclusion: definition and measurement. We defined financial inclusion as vulnerable households’ ability to access useful and affordable financial products and services, such as transactions, savings, credit, and insurance (CBN, 2012; Wentzel et al., 2016). These financial products were measured in this study similarly to the main approaches adopted in the macroeconomics literature. For instance, one strand of the literature measures financial access as a proxy of the proportion of individuals with transaction accounts or the size of the banked population (Ibrahim, 2014; Park & Mercado, 2015). Moreover, an equally weighted five-pillar approach, consistent with the contextual objective of financial inclusion (transaction, smoothing, and resilience), was adopted as a methodological strategy for measuring FI. Therefore, we used variables that determine whether a household is accessing either micro-credit, demand deposits, time deposits, or insurance from the formal financial market, as highlighted in Table 1. Moreover, the standardisation of the components of FI  follows a similar strategy to the United Nation Development Program’s (UNDP) (2014) computation of the multidimensional poverty index (MPI).

Given that the improvement in welfare of the rural populace is one of the major desirable outcomes of financial inclusion, the choice of indicators of welfare remains central to the analysis of this study. In this sense, we followed two strands of research that used the logarithm of per capita expenditure (see Ibrahim, 2014; Dzanku, 2015) as a simplified measure of welfare and other strands of literature (Kirsten, 2012; Dzanku, 2015; Gautam & Andersen, 2016) that advocated the construction of a self-reported well-being index (SWI) using the multidimensionality of various factors that combines the non-money metrics and monetary indicators of the living standard.

While measuring welfare via the SWI, people are often asked to not only free-list the components that characterise quality of life or well-being, but also to classify the listed components into three broad categories consisting of housing, consumption expenditure and wealth, as presented in Table 2. Then, 15 household-specific indicators associated with these components are selected according to their functional importance in the local context. While the different basic household facilities are accumulated or built over a relatively longer time, they are a better reflection of well-being than other indicators such as income, which fluctuates within shorter time periods (Aliero & Ibrahim, 2012; Gautam & Andersen, 2016). Out of all the 15 indicators, eight are used in measuring the components of housing including possession of basic household assets as well as the status of sanitation. The first set of thematic questions on this component relate to the household’s experience of having certain basic facilities that add quality to their life. For instance, questions are asked to ascertain ownership of electronics, furniture and electricity. Where the response is affirmative, follow-up questions are then asked to list the type of item owned (e.g. radio, television, video, the source of the electricity: a generating set or national grid, etc.). Moreover, the second component measures the aspect of food security and the medical expenditure of the households. To avoid the potential effects of inflation, these food and medical expenses are measured in Nigerian naira and then converted to US dollars. Lastly, wealth measures the stability of subsistence as the ability to afford social functions such as weddings, naming ceremonies, festival and mortuary rites are determined by the level of wealth that the household possesses. In this sense, a household with a high level of well-being has adequate wealth, particularly cash and formal savings and livestock, so that they can be used against contingencies as well as to stabilise the desired level of subsistence (Hassan & Birungi, 2011; Kirsten, 2012; Dzanku, 2015). We then calculate the SWI by adopting the MPI methodology that assumes equal weight, and a linear and additive relationship among the various aforementioned components. This computational strategy is used because it is proved to be the most efficient and consistent new development index that focuses on microeconomic indicators (Dzanku, 2015; Gautam & Andersen, 2016).

Table 2. Variables used for constructing a well-being index.

Component (weight)	Sub-component (weight)	Indicators (weight)	Measurement
Housing (1/3)	Basic household assets (1/6)	Owned electronics (1/24)	1 = yes; 0 = otherwise
		Furnished (1/24)	1 = yes; 0 = otherwise
		Has electricity (1/24)	1 = yes; 0 = otherwise
		Has separate living room (1/24)	1 = yes; 0 = otherwise
	Sanitation (1/6)	Household-specific access to safe drinking water (1/24)	1 = yes; 0 = otherwise
		Household-specific access to kitchen (1/24)	1 = yes; 0 = otherwise
		Household-specific access to latrine (1/24)	1 = yes; 0 = otherwise
		Has first aid box (1/24)	1 = yes; 0 = otherwise
Consumption expenditure (1/3)	Food consumption (1/6)	Average weekly food expenditure (1/6)	Converted from Nigerian Naira to USD
	Medical expenditure (1/6)	Average weekly medical expenditure (1/6)	Valued in USD
Wealth (1/3)	Savings (1/6)	Cash and formal savings (1/12)	Valued in USD
		Livestock (1/12)	Total livestock unit per capita (USD)
	Other assets (1/6)	Landed assets (non-farm) (1/12)	Valued in USD
		Acres of farm (1/12)	Valued in USD

3. Data and descriptive statistics

Data were collected in phases, between October 2014 and September 2015. The study adopted a stratified multi-stage sampling procedure, which involved simultaneous determination of the survey area and respondents. We purposely selected villages with existing formal financial institutions, so that the decision of households not to use financial services was not related to a lack of banking outreach. Moreover, a random sampling process was applied in selecting the survey areas within each identified village. As a result, between 170 and 180 households (for a sample size of 1 750 respondents) were systematically selected from 10 villages in northern Nigeria, with an estimated adult population of 250–300. Furthermore, households, as opposed to household heads, were used as the unit of analysis.

The data seemingly provides the dynamics of financial inclusion and relevant livelihood activities. The survey questionnaire was designed to solicit a considerable amount of details on demographics, income sources, access to formal financial services, and relevant well-being indicators. However, out of the 1 750 surveyed households, 68 (or 4%) were unable to provide complete information on some indicators for the well-being index, either by omission or commission. To avoid bias, 4% of the data were adjusted for the missing data by standardising the components of the well-being score. This accounted for slight differences in the data used for the proxy of welfare: 1 750 and 1 682 observations for the per capita expenditure and well-being index, respectively.

The stratified descriptive data show that, despite moderate overall financial inclusion scores (approximately 28%, as per Table 3), financially-included households have relatively more favourable statistics in terms of welfare and overall human capital. This leads to the central conjectures this study addresses systematically: the welfare effect of financial inclusion > 0, financial inclusion effect on various income distribution > 0, and income (or welfare) convergence Q_50–Q_10 < 0 and Q_90–Q_50 < 0.

Table 3. Descriptive statistics of households surveyed.

Variables	Overall	Financially included	Financially excluded	t-value
Gender (1 = male) (µ)	0.679	0.277	0.723	−13.982**
Age (µ)	42.510	41.321	43.584	−1.630
Household size (µ)	12.163	10.442	13.643	−8.543**
Literacy (µ)	2.773	3.011	2.329	4.732**
Per capita expenditure (µ)	2.504	2.601	2.454	3.442**
Well-being score	0.412	0.583	0.387	6.221**
Financial inclusion (µ)	0.276	1.000	0.000	–
N	1 750	483	1 276

**denotes significance at the 0.01 level.

4. Results and discussion

4.1. Regression results

The estimated results of Equation (3) are presented in Table 4. Iteratively, the objective here is to establish the major determinants of household welfare. Models (1) and (2) are the overall estimates using per capita expenditure and the well-being index, respectively. Models (1a) and (2a) were computed with data from financially included households; while Models (1b) and (2b) show the estimated coefficients of financially excluded households. It is important to note that most coefficients are consistent with a priori expectations. The variable of concern is financial inclusion, and the results show that it exerts a strong positive impact on household welfare. This supports the findings of previous studies (Chibba, 2009; Kirsten, 2012; Ibrahim, 2014; Park & Mercado 2015).

Table 4. Determinants of household welfare.

Variables	Per capita expenditure (log)						Well-being index
	Model (1)		Model (1a)		Model (1b)		Model (2)		Model (2a)		Model (2b)
	Coef.	Std. error	Coef.	Std. error	Coef.	Std. error	Coef.	Std. error	Coef.	Std. error	Coef.	Std. error
FI			0.541**	0.051					0.311**	0.099
Household’s income	0.327**	0.120	0.423**	0.083	0.641**	0.121	0.453**	0.072	0.443**	0.072	0.238*	0.115
Household’s age	0.228**	0.084	0.319*	0.128	0.153**	0.028	0.214**	0.082	0.201**	0.025	0.120	0.125
Household’s gender	−0.126	0.156	−0.161*	0.072	−0.115*	0.022	−0.204**	0.048	−0.211**	0.055	−0.098	0.063
Household size	0.212**	0.042	0.383**	0.064	−0.149**	0.044	−0.250**	0.062	0.410**	0.084	−0.212**	0.074
Literacy	0.433*	0.195	0.488**	0.057	0.302**	0.031	0.511**	0.074	0.308**	0.062	0.451**	0.042
R^2	0.41		0.39		0.35		0.45		0.42		0.36
N	1750		483		1276		1682		463		1219

Notes: Models (1) and (2) are overall models, using consumption expenditure and composite welfare index, respectively. Similarly, Models (1a) and (2a) were computed with data on financially included households, while Models (1b) and (2b) estimate the coefficients of financially excluded households.

* and ** denote significance at the 0.05 and 0.01 levels, respectively.

To show whether the welfare differences between financially included and deprived households are driven by financial inclusion, we present the quantile regression result in Table 5, highlighting the welfare differences within the financially included (or excluded) (i.e. within group difference) and those between the financially included and excluded (i.e. between group difference). The effect of financial inclusion is better welfare, with the rapid increase of regression coefficients across each quantile, when compared to a slower effect for financially deprived households, particularly at lower quantiles. Interestingly, despite pre-existing inequalities within the two groups of households, the welfare disparity at the lower quantiles distribution [0.249(Q_50–Q_10)] is much larger compared to the relatively narrower disparity at higher quantiles [0.096(Q_90–Q_50)] for financially included households. This welfare inequality is consistent, even when using the well-being index as a dependent variable. A key finding is that financial inclusion could lower welfare inequalities between middle- and high-income households, while inequalities within low-income households could exacerbate. Conversely, inequality is higher at higher quantiles [0.149(Q_90–Q_50)] than in lower ones [0.062(Q_50–Q_10)] for deprived households.

Table 5. Estimate of the welfare differences between financially included and deprived households.

	Dependent variable: Per capita expenditure (log)			Within group diff.
Status	Q_10	Q_50	Q_90	Q_50–Q_10	Q_90–Q_50
Financially included	0.093** (4.39)	0.342** (5.61)	0.437** (4.11)	0.249	0.096
Financially deprived	0.032** (4.10)	0.094** (5.45)	0.243* (2.12)	0.062	0.149
Between group diff.	0.061	0.248	0.194
	Dependent variable: Well-being index
Financially included	0.061* (2.25)	0.193** (5.23)	0.272** (4.47)	0.132	0.079
Financially deprived	0.058** (4.34)	0.140** (3.51)	0.180** (5.41)	0.082	0.040
Between group diff.	0.003	0.053	0.092

* and ** denote significance at the 0.05 and 0.01 levels, respectively. t-values are between parentheses.

4.2. Paths of welfare enhancement

Following the findings in the preceding section, we trace the paths to welfare enhancement of livelihood activities, as this would identify the economic activity that exerts a greater impact on the reduction of welfare disparity among the various household income categories. The results in Table 6 show a strong welfare effect from business profits, off-farm wages, and earnings from exploiting environmental resources. Interestingly, a marginal inequality exists between lower and higher quantiles for the households engaging in these activities. The negative sign of within group differences in business profit for financially included households at lower [−0.068(Q_50–Q_10)] and higher quantiles [−0.114(Q_90–Q_50)] indicates that trade has not only the highest welfare effect, but also exerts a stronger impact for decreasing welfare inequality. The business profit for financially deprived households shows inequality across all quantiles. Moreover, earnings from livestock production have a negative inequality effect on welfare across all quantiles, but the impact is relatively weak. A higher incidence of livestock theft by cattle rustlers is the possible reason for having a weaker livestock earning-welfare nexus.

Table 6. Paths to welfare enhancement.

	Dependent variable: Per capita expenditure (log)
	Financially included					Financially deprived
				Within group diff.					Within group diff.
Activities	Q_10	Q_50	Q_90	Q_50–Q_10	Q_90–Q_50	Q_10	Q_50	Q_90	Q_50–Q_10	Q_90–Q_50
Profit from own business	0.453** (5.39)	0.385** (5.14)	0.271** (3.64)	−0.068	−0.114	0.156** (4.31)	0.262** (5.43)	0.283** (4.63)	0.106	0.021
Off-farm wages	0.032* (2.10)	0.201** (6.83)	0.387** (4.21)	0.169	0.186	0.051** (3.83)	0.152** (3.34)	0.189* (2.12)	0.101	0.037
Farm wages	−0.341** (−4.15)	0.121** (6.22)	0.050** (5.20)	0.462	−0.071	0.021** (3.23)	0.094** (5.23)	0.117** (5.13)	0.073	0.023
Earnings from crop production	0.314** (5.62)	0.182** (5.51)	0.081** (5.12)	−0.132	0.101	0.154** (5.51)	0.162** (5.11)	0.271** (6.43)	0.008	0.109
Earnings from livestock rearing	−0.091** (−3.15)	−0.108 (−1.51)	−0.122* (−2.23)	−0.017	−0.014	0.008 (1.21)	−0.018* (−1.98)	−0.091* (−2.16)	−0.026	−0.073
Environmental resource extraction	0.361** (4.74)	0.263** (5.23)	0.185* (2.23)	−0.098	−0.078	0.251** (5.23)	0.234** (5.41)	0.211** (7.56)	−0.017	−0.023
Area_dummy	−0.213** (−3.54)	−0.164* (−2.11)	−0.180* (−2.32)	0.049	−0.016	−0.142* (−2.13)	−0.148** (−3.23)	−0.151** (−3.56)	−0.006	−0.003
R^2	33	35	32			33	34	34
Dependent variable: Well-being index
Profit from own business	0.321** (4.72)	0.261** (3.55)	0.352** (5.14)	−0.068	0.091	0.085** (3.14)	0.162** (5.12)	0.173** (3.61)	0.077	0.011
Off-farm wages	0.096** (5.11)	0.151** (3.61)	0.184** (3.24)	0.055	0.033	0.062** (3.83)	0.092* (2.13)	0.159** (3.52)	0.03	0.067
Farm wages	−0.144** (−5.14)	−0.133** (−4.12)	0.082** (3.54)	0.011	0.215	0.021** (6.22)	0.080** (5.23)	0.102** (4.16)	0.059	0.022
Earnings from crop production	0.202** (3.21)	0.152** (4.81)	0.111* (2.12)	−0.050	−0.041	0.082** (3.51)	0.102** (6.16)	0.151** (4.44)	0.002	0.049
Earnings from livestock rearing	−0.032** (−4.27)	−0.074* (−2.11)	−0.092** (−4.32)	−0.042	−0.018	−0.048** (−4.51)	−0.083** (−3.43)	−0.098** (−4.64)	−0.035	−0.015
Environmental resource extraction	0.183* (2.12)	0.141** (4.31)	0.102** (4.21)	−0.042	−0.039	0.063** (5.23)	0.082** (3.41)	0.091** (4.56)	0.019	0.009
Area_dummy	−0.206** (−4.41)	−0.208** (−3.18)	−0.212* (−2.12)	−0.002	−0.004	−0.142** (−6.22)	−0.142* (−2.11)	−0.125** (−4.75)	0.000	0.017
R^2	31	30	30			32	32	30

* and ** denote significance at the 0.05 and 0.01 levels, respectively. t-values are between parentheses.

Another important path to welfare augmentation is earnings from crop production. The gap in the well-being index of financially included households at lower [−0.005(Q_50–Q_10)] and higher quantiles [−0.041(Q_90–Q_50)] is minimal, indicating that earnings from crop production could ameliorate welfare disparities. This reinforces the calls for using agro-credit as an alternative means of reducing poverty and welfare inequalities. Thus, the SSA, particularly Nigeria needs to intensify and widen the agro-credit programme to cover the entire agricultural sector, so as to achieve higher overall standards of living.

4.3. Robustness tests

As robustness analysis, we estimate the OLS model of the relationship between per capita income, FI, and welfare. The results, presented in Table 7, show a strong positive impact of financial inclusion on per capita income. Similarly, the estimated quantile results from Q_10–Q_50 show smaller coefficients of FI, with greater inequality across the lower quantiles and marginally stationary coefficients at higher quantiles, especially from Q_60 to Q_90. Moreover, when controlling for an area of residence (area dummy), the results show the per capita income of the households residing in conflict-prone villages to be 31% lower than for households residing in the less hostile areas. If the argument of internal dualism as a breeding ground for organised crime holds (Gurr, 2005; Agbiboa, 2013), then the need for inequality eradication driven by a financial inclusion policy becomes more pronounced, in order to decrease the perceived income disparity between rich and poor households.

Table 7. Household income, financial inclusion, and welfare.

	Dependent variable: Per capita income (log)
Variables	OLS	Q_10	Q_20	Q_30	Q_40	Q_50	Q_60	Q_70	Q_80	Q_90
Well-being index	0.443** (6.21)	0.026* (2.13)	0.047** (6.53)	0.062* (2.50)	0.083** (6.48)	0.113** (5.82)	0.132** (4.32)	0.140** (5.53)	0.202** (6.31)	0.261** (8.23)
FI	0.311** (4.82)	0.105** (5.41)	0.143** (6.15)	0.210** (4.52)	0.240** (6.43)	0.272** (4.31)	0.285** (4.73)	0.301** (3.52)	0.308** (5.53)	0.310** (4.21)
Household age	0.210* (3.82)	0.214** (3.35)	0.180** (4.12)	0.162** (4.82)	0.140* (2.12)	0.123 (1.27)	0.105** (5.23)	0.098** (4.16)	0.072 (1.15)	0.053 (1.20)
Household gender	−0.135** (−4.23)	−0.182** (−4.62)	−0.173** (−5.41)	−0.156** (−6.42)	−0.142** (−4.40)	−0.131** (−3.52)	−0.122** (−5.41)	−0.114* (−2.41)	−0.094* (−2.25)	−0.081** (−5.11)
Household size	0.224** (5.51)	0.241** (5.15)	0.239** (3.23)	0.227** (4.41)	0.213* (2.19)	0.183* (2.11)	0.163 (1.43)	0.148 (1.51)	0.131* (2.23)	0.120 (1.21)
Literacy	0.412** (4.13)	0.161** (4.41)	0.180** (3.22)	0.185** (5.03)	0.202** (4.14)	0.243* (2.23)	0.264* (2.14)	0.282** (4.56)	0.308** (5.53)	0.356* (2.21)
Area_dummy	−0.301** (−3.12)	−0.183** (−5.51)	−0.173** (−3.32)	−0.163** (−4.23)	−0.160** (−5.24)	−0.152** (−4.10)	−0.144* (−2.21)	−0.135** (−5.16)	−0.122** (−3.45)	−0.111** (−4.18)
R^2	0.43	0.35	0.35	0.34	0.36	0.34	0.35	0.36	0.34	0.36
N		483

* and ** denote significance at the 0.05 and 0.01 levels, respectively. t-values are between parentheses.

Following relevant literature (Caliendo & Kopeinig, 2008; Park & Mercado, 2015), we implemented PSM with multiple matching algorithms as a robustness exercise. The results in Table 8 show that the coefficient of average treatment effect on the treated (ATT) is closer to the coefficients of OLS, with the same level of significance for all the estimated models. For instance, using nearest to neighbour as the matching algorithm yields an ATT of 0.317 (using the FI estimator), which means that welfare of the financially included households is approximately 32% higher than that of financially deprived ones.

Table 8. Propensity score matching.

Estimator	Baseline result	Matching methods
	OLS	Nearest neighbour	Radius	Stratification	Kernel	Local linear regression
FI	0.311** (0.063)	0.317** (0.083)	0.331** (0.076)	0.321** (0.083)	0.347** (0.068)	0.325** (0.062)
Well-being index	0.443** (0.082)	0.451** (0.091)	0.442** (0.065)	0.434** (0.057)	0.423** (0.047)	0.441** (0.064)

** denotes significance at the 0.01 level. Standard errors are between parentheses.

Further to the robustness analysis based on Machado and Mata (2005), the counterfactual decomposition test highlighted in Table 9 is consistent with our previous findings. The results show that, from the 10th to the 40th quantiles, the proportion of the coefficients effect is higher than that of the characteristics effect, suggesting that financial inclusion, rather than household characteristics, is contributing to the income inequalities for low-income households. However, inequality at the upper-middle quantiles (50th until 70th) is explained by the characteristics effect, particularly literacy, gender, and area of residency.

Table 9. Counterfactual decomposition.

	Q_10	Q_20	Q_30	Q_40	Q_50	Q_60	Q_70	Q_80	Q_90
Coefficient effect	68.54	66.21	64.75	55.01	47.83	46.11	48.15	51.31	51.88
Characteristics effect	31.46	33.79	35.25	44.99	52.17	53.89	51.85	48.69	48.12
Total	100

Note: Q_10 to Q_90 indicate quantiles from 10 to 90.

A major difference from the previous studies lies in the ambiguities associated with the extent to which financial inclusion is affecting welfare. On the one hand, some studies have claimed that the benefit of financial inclusivity is welfare convergence across all households (Chibba, 2009; Wentzel et al., 2016), while others revealed that financial inclusion promotes inequality (Ray, 1998; Fafchamps, 2004). While our findings do not entirely dismiss the possibility of either claim, they settle the debate by identifying that welfare equality is only possible for middle- and high-income households, while pre-existing inequality within the low-income households could exacerbate.

5. Conclusions and policy implications

This study examines the finance-welfare nexus by constructing a multi-variable financial inclusion index regressed against household welfare. As expected, OLS shows a strong positive effect of financial inclusion on household welfare. On the other hand, quantile regression reveals middle- and high-income households benefitted relatively more from financial inclusion, compared to lower income households. Therefore, we argue that for financial inclusion to alleviate welfare inequality and ensure income convergence, rural financial markets must be redesigned to allow wider access to credit, specifically for low-income and vulnerable households. First, there is an urgent need for policy reversal to reduce the exorbitant interest rate and other exploitative hidden charges that low-income households could not possibly afford while transacting with rural deposit financial institutions (RDFIs). The practices of charging a high rate of interest on loan facilities and crediting depositor’s account with minimal interest rate can compromise the financial inclusion objective of alleviating poverty. Moreover, there is the need to move away from the one-size-fits-all to a more diversified customer-centric model where financial services would be simultaneously provided on interest-based vis-à-vis interest-free financial windows. Diversification of financial services is thus a crucial step towards exiting the financial exclusion trap. In this way, interest-free development finance institutions could play a dedicated role in closing financial gaps by facilitating the interest-free financial intermediation.

This study identified a strong welfare augmentation effect from informal livelihood strategies, such as environmental resource extraction, and crop and livestock production. While these activities do not seem to be credit attractive due to their associated uncertainty, especially the rain-fed cropping and pastoral livestock production. This implies that special effort is needed for the RDFIs to collect information and develop their own risk assessment tools in order to go beyond the perceived risk syndrome (Giordano & Ruiters, 2016). With the recent emphasis on reinvigorating the agricultural sector in SSA that has subsequently improved the credit rating of farmers, RDFIs need to draw on informal financial institutions while developing their risk assessment tools so as to reduce credit rigidities in the informal and semi-formal sectors of the economy. While we have allowed space for future research to establish the causality between financial inclusion and income diversification, we acknowledge that although PSM has been applied in controlling potential endogeneity, this problem cannot be completely ruled out.

Disclosure statement

No potential conflict of interest was reported by the author.

Acknowledgments

We would like to commend the editor and anonymous reviewers for providing valuable suggestions and insightful and constructive comments on an earlier version of this article. We are also thankful to HM Aliero for his helpful comments. Data, codes and DO-files used in this article are available from the authors upon request.

ORCID

Saifullahi Sani Ibrahim http://orcid.org/0000-0001-8708-4056

Behiye Cavusoglu http://orcid.org/0000-0003-0561-4514

Additional information

Funding

Data collection was partly supported by the Tertiary Education Trust Fund and the Academic Staff Training & Development fellowship, awarded to the corresponding author through the Federal University Dutsin-ma (FUDMA) Institutional Grant [grant number TETFUND/IBR 165].