Risk Aversion, Self-Control, Commitment Savings Device and Benchmark-Defined Undersaving Among Nano Enterprises in Urban Slums: A Logistic Regression Approach

Abstract

Low-income individuals are unlikely to save relatively large sums on a regular basis; however, many still fall short of even the modest threshold required for long-term financial security. This study examines the determinants of benchmark-defined undersaving among retail e-payment agents (REAs) operating in the urban slums of Lagos, Nigeria. We use a contingent valuation survey, descriptive analysis, and logistic regression to examine how selected behavioural and demographic factors, alongside a 60-day experimental intervention—the Programmed Microsaving Scheme (PMSS), a hard daily commitment savings device—affect the likelihood of undersaving, defined as saving less than 12% of each REA’s average daily income. While the PMSS appears to have contributed to improvements in post-treatment saving participation and performance among REAs, it did not significantly increase the likelihood of reaching or exceeding the benchmark savings threshold. Consistent with this, average daily income, age, gender, marital status, education, and religion are statistically insignificant predictors of benchmark-defined undersaving. In contrast, self-control, measured using a literature-validated instrument, exhibits a statistically significant negative association with benchmark-defined undersaving, indicating that higher self-control reduces the likelihood of failing to meet the benchmark. Measured risk aversion similarly shows no significant association. Notably, this study introduces a novel 60-day PMSS, co-designed with REAs and neobanks to accommodate daily income savings—a characteristic of the informal sector largely overlooked in the literature on commitment savings devices. From a policy perspective, the findings suggest that while short-horizon commitment devices (such as the 60-day PMSS) and financial literacy are associated with improvements in microsavings among low-income daily earners, achieving benchmark-level saving might require longer-term and more adaptive mechanisms that address income volatility and mitigate other inherent risks.

1. Introduction

In developing countries, there is a long-standing and growing recognition that low-income individuals, households, and micro or nano enterprises (MINAEs) benefit substantially from barrier-free access to formal microsaving (Kelley & Williamson, 1968; Adams, 1978; Vogel, 1984; Zeller & Sharma, 2000; Armendáriz de Aghion & Morduch, 2005; Hulme et al., 2009; Gunhild, 2010; Delgado et al., 2015; Liu, 2018; Di Giannatale & Roa, 2019; Arora, 2022; Klapper et al., 2025). In response, many financial inclusion (FI) initiatives—supported by governments, development agencies, donors, and policymakers—and national financial inclusion strategies (NFISs) implemented by sovereign members of the Alliance for Financial Inclusion (AFI) have successfully prioritised access to formal microsavings through diverse providers. As evidence of this, global account ownership at banks, credit unions, microfinance institutions (MFIs), and mobile money operators rose from 51% of adults in 2011 to 79% in 2024 (Klapper et al., 2025), while in developing countries, it increased from 63% in 2017 to 71% in 2021 (Demirgüç-Kunt et al., 2022). Yet, many FI stakeholders fail to recognise that formal account opening is merely the first step toward meaningful financial inclusion (Sen & De, 2018); the more significant challenge lies in fostering active, optimal, and sustained account use to enhance welfare (Banerjee & Duflo, 2011), whether for short-term gains, pensions, or other long-term investment goals. This study argues that policymakers, formal microsaving providers (FMSPs), donors, development agencies, governments, and MINAEs themselves must act urgently and collaboratively to address undersaving—a critical issue largely neglected in current NFISs and research.

Karlan et al. (2014) define undersaving as “a lower level of savings than one would have in a world with perfect markets (perfect information, zero transaction costs, and perfect competition among financial institutions) and fully attentive, fully rational, and fully consistent decision-making”. Put differently, the sub-optimal saving levels observed among low- or high-income individuals stem from behavioural factors, information asymmetry, transaction costs, and broader market inefficiencies, imperfections, frictions, and failures (MIIFFs). Literature reviews by Karlan et al. (2014), Di Giannatale and Roa (2019), and Osifodunrin and Lopes (2022) identify numerous determinants of microsaving behaviour, saving rates, savings adoption, undersaving, and related sub-constructs of formal microsaving development (FMSD). These factors can be grouped as follows:

• Economic factors: Inflation, (real) interest rates, GDP, foreign capital, economic fluctuation, labour formality, (un)employment, financial liberalisation (proxied by access to microcredit), and transaction costs.
• Social factors: Social capital, culture, networks, financial education, government transfers, remittances, subsidies, distrust in FMSPs, and related constraints.
• Institutional and technological factors: Institutional quality, technologies (e.g., mobile money), regulatory barriers, FMSPs’ innovations, and institutional impact.
• Geographic and demographic factors: Distance-related barriers, household/individual income of low-income groups, home ownership, education, gender, age, marital status, income growth, and dependency ratio.
• Behavioural and psychological factors: Biases affecting microsaving and use of soft or hard commitment devices.

A comparative review of these categories against sub-constructs of FMSD shows that behavioural factors remain the least investigated (as independent variables), while undersaving (as a dependent variable or sub-construct of FMSD) has been largely overlooked—particularly for MINAEs in developing countries. This study therefore commits to the following:

(a)

Introduce the concept of benchmark-defined undersaving, which refers to a situation where actual saving behaviour falls below a predefined, normative, stakeholder-validated, or policy-reflective benchmark. This benchmark determines whether saving is “adequate” or “inadequate” for preventing or mitigating the severity of future economic hardship or old-age poverty, as emphasised by Unnikrishnan and Imai (2020) and Barrientos et al. (2003). This non-trivial concept may be of particular significance in less-welfarist jurisdictions where governments are less often perceived as the rescuer of last resort and where citizens must necessarily make private pension arrangements to forestall or mitigate future economic hardship (Boto-García et al., 2022; Kośny et al., 2024). In addition, the concept addresses a fundamental policy-relevant question: whether individuals are saving sufficiently to protect themselves against future economic hardship, rather than just saving. By anchoring assessment to a clear, socially agreed benchmark, it enables policymakers and researchers to identify at-risk populations, improve intervention targeting, and evaluate whether financial inclusion and saving initiatives translate into real gains in economic security. For poorer households, the concept recognises saving effort while revealing structural constraints, showing that regular saving may still be inadequate due to low or unstable incomes. As such, benchmark-defined undersaving functions as a preventive, dignity-preserving indicator, supporting early policy action and fairer intervention design while reducing reliance on ex post emergency assistance or government bail-outs. As revealed and further explained in Section 3 and other sections of the current paper, benchmark-defined undersaving occurs when REAs save less than 12% of their average daily income.

(b)

Investigate the empirical relationship between the measured self-control of retail e-payment agents (REAs) and their undersaving behaviour (operationalised with the introduced benchmark-defined undersaving). REAs, also known as branchless banking agents (Lyman et al., 2006, 2008; Ledgerwood et al., 2013; M. A. Ashraf, 2022), are MINAEs that support the FI journeys of low-income individuals and other MINAEs. They operate as agents for deposit money banks (DMBs), neobanks, mobile money operators (MMOs), and payment service providers (PSPs), delivering technology-enabled microsaving, deposits and withdrawals, transfers, bill payments, and related services, for which they earn regulated, transaction-based commissions that constitute their primary source of daily income. The International Finance Corporation defines micro-enterprises as businesses with fewer than ten employees, total assets under USD 100,000, and annual sales below USD 100,000 (International Finance Corporation [IFC], 2013). In this study, REAs are more accurately classified as nano enterprises, since promoters typically operate as sole employees and the asset base—mainly a point of sale (POS) terminal and minimal cash reserves for withdrawals—rarely exceeds USD 1000, as confirmed by the field survey. This study centres on REAs—a critical yet overlooked MINAE subgroup (Osifodunrin & Lopes, 2022)—whose undersaving tendencies are noted. As FI crusaders or agents, policies and studies targeting them may also (in)directly influence the low-income groups they serve.

(c)

Examine how measured risk aversion—linked to REAs’ operational risks and as previously assessed by Dohmen et al. (2011) and Hardeweg et al. (2013)—affects their benchmark-defined undersaving.

(d)

Prior studies (Dagnelie & Lemay-Boucher, 2012; N. Ashraf et al., 2006) show that commitment devices, such as the Programmed Microsaving Scheme (PMSS) introduced in this study, help low-income groups save more by mitigating self-control problems. This study examines the PMSS treatment’s effect on REAs’ benchmark-defined undersaving.

(e)

Finally, the study assesses how income and demographic factors (mainly as control variables) shape REAs’ benchmark-defined undersaving.

The justification for this study rests on the benefits of increased microsaving or the mitigation of undersaving. Horioka and Watanabe (1997) classify saving motives as life-cycle, precautionary, and bequest. Formal microsaving plays consumption-smoothing and insurance roles for low-income groups, particularly in cushioning income shocks, hardship, or emergencies (Hulme et al., 2009; Steinert et al., 2022). Beyond supporting the accumulation of capital and the enhancement of living conditions, microsaving fosters financial discipline, facilitates access to default-free microloans, and promotes financial independence. Ledgerwood (1999) emphasises its role as a “collateral substitute”, particularly when micro-lenders require compulsory pre-loan savings. It also enhances provider–customer relationships via Know-Your-Customer (KYC) processes, deposit histories, and opportunities for cross-selling other financial products.

Microsaving also lowers the cost of funds for microcredit compared to other sources (Rutherford, 2001; Portocarrero et al., 2006; World Economic Forum [WEF], 2008; Wesley & Palomas, 2010; Demirgüç-Kunt et al., 2014). This enables MFIs to extend more loans at lower interest rates while balancing the risks of mismatches between deposits and lending. Time-bound deposits, being less volatile, provide more predictable liquidity management. Moreover, MFIs funded by microsaving often enjoy stronger community trust and social capital, discouraging loan defaults as borrowers recognise that they are drawing from communal resources. Governments also benefit since optimal microsaving accumulation aligns with poverty reduction and related political objectives (Deaton, 2010; Martin et al., 2013; Doering & McNeill, 2020).

Findings from this study’s baseline survey confirm that undersaving is real and severe. All surveyed REAs reported chronic undersaving, with daily savings below the group-validated and policy-reflective threshold of 12% of average daily income. Thus, undersaving is not abstract but a concrete developmental challenge undermining low-income households.

Despite extensive research, no previous study has jointly examined risk aversion, self-control, and the role of commitment savings devices (CSDs) in influencing undersaving among often-overlooked MINAEs. This study therefore advances the literature and informs policymakers about which factors significantly affect undersaving or benchmark-defined undersaving. Still, income, demographics, and behavioural traits alone may not fully explain REAs’ saving behaviour. As noted by Koutsoyiannis (1977) and Wulandari (2022), unobserved influences (such as the unique financial position of each REA, socio-cultural pressures, and others) can be captured within the econometric random error term (ε), as modelled in Section 4.

Thaler and Benartzi (2004) further show that in countries shifting from defined-benefit to defined-contribution pension plans, mandatory saving rates often mirror predicted life-cycle savings. Individuals saving far below such benchmarks—particularly informal or semi-formal workers like REAs—may suffer from self-control lapses, lack of awareness, or irrational choices, requiring targeted interventions. The behavioural mechanisms deployed in this study, such as the PMSS and the comprehensive microsaving enlightenment programme (CMEP), align with recommendations from Thaler and Benartzi (2004), N. Ashraf et al. (2006), and Karlan et al. (2014) on reshaping microsaving decision-making. Examples of policy-recommended pension saving rates include 18.5% of a worker’s monthly salary in Ghana (Government of the Republic of Ghana [GRG], 2008) and 18% of total monthly emoluments in Nigeria (Federal Government of Nigeria [FGN], 2014). Focusing on Nigeria, this study initially adopted the minimum mandatory savings rate stipulated in Nigeria’s national pension policy for the formal sector, which is 18%—with employees contributing at least 8% of monthly emoluments and employers contributing a minimum of 10% on their behalf (Federal Government of Nigeria [FGN], 2014).

The remainder of this paper is organised as follows. Section 2 reviews the relevant literature. Section 3 details the data and methodological approach, while Section 4 presents the empirical results. Section 5 concludes with a discussion of the study’s limitations, and Section 6 highlights the key policy implications.

2. Brief Literature Review

This section first outlines the typical and conceptual explanations of undersaving behaviour within the context of low-income individuals and groups. It then discusses the factors hypothesised to have statistically significant effects on undersaving.

2.1. Undersaving and the Low-Income Groups

In recent decades, many countries have prioritised pension privatisation and cutback on welfarism, making contributory pensions and private (micro)saving critical for sustaining post-retirement living standards (Boto-García et al., 2022; Kośny et al., 2024). Yet, low-income individuals (particularly in developing countries) grossly undersave. This behaviour is reflected in several ways:

(i)

They often fail to set or adhere to microsaving goals, showing weak self-discipline, impulsive spending, and a preference for short-term gratification over long-term stability. As a result, they typically lack emergency savings, consistent with the present bias explanations and behavioural life-cycle hypothesis (Shefrin & Thaler, 1988).

(ii)

Many face uncertainty and hopelessness that discourage future-oriented actions such as saving or investing (Banerjee & Duflo, 2011). Conversely, some persistently undersave due to “optimism bias”—expecting future windfalls or higher earnings that rarely materialise—thereby distorting rational saving (Sharot, 2012).

(iii)

Even under worsening financial conditions, they often fail to adjust spending habits and, in extreme cases, resort unsustainably to overpriced microcredit.

(iv)

Among MINAEs in developing countries, undersaving often reflects the weak saving preferences of promoters. It also reflects deficient knowledge in most aspects of institutional saving mechanism, discipline, and governance structures. Consequently, their microsaving—an investment in their future—is consistently undermined. According to Qureshi (1983), Chen et al. (2017), and Ghosh and Nath (2023), in more structured corporate entities, (under)saving would often reflect the degree to which firms retain, distribute, or expend their profits.

(v)

More broadly, undersaving overlaps conceptually and empirically with “lack of microsaving”, “overspending”, “compulsive consumption”, “overreliance on microcredit”, and “chronic indebtedness”, which may serve as proxies for sub-optimal saving behaviour (Baumeister, 2002; Kamleitner et al., 2012; Achtziger et al., 2015).

These undersaving behaviours may stem from poor self-control, risk aversion, limited knowledge, or lack of awareness about inherent socio-economic risks and the importance of microsaving. Accordingly, this study’s focus on undersaving extends broadly to the diverse benefits and applications of microsaving—whether in micropension, targeted savings, asset acquisition, bequest, micro-investment, consumption smoothing, precautionary savings, or microinsurance.

2.2. Self-Control and Undersaving

Self-control is the ability of economic agents to regulate behaviour and resist immediate gratification in pursuit of long-term goals. This regulation is often difficult, with many admitting to saving below their own optimal thresholds (Bernheim, 1994). Empirical evidence shows that higher self-control improves saving behaviour (Brandstatter & Guth, 2000; Webley & Nyhus, 2001). Accordingly, this study hypothesises an inverse relationship: as self-control increases, undersaving declines across income groups.

Earlier research assumed or suggested that employees in the formal sector and higher-income earners exhibit greater saving discipline than informal and low-income workers, largely due to access to formal pensions, automatic payroll deductions, and other structured saving mechanisms (Collins et al., 2009; Banerjee & Duflo, 2011). However, the rapid expansion of mobile money and digital financial platforms—even into remote, low-income communities (particularly after COVID-19)—calls for a re-examination of these assumptions. While access to digital and mobile-enabled financial tools has expanded, adoption is shaped by three key factors: awareness, financial sophistication, and self-discipline. Despite widespread mobile phone ownership, many adults remain unbanked; in some South Asian countries, more than half of unbanked adults still own mobile phones (Demirgüç-Kunt et al., 2022).

Although no study (in peer-reviewed journals) directly examines self-control and undersaving among MINAEs in developing countries, related evidence is instructive. Ameriks et al. (2007), Choi et al. (2017), and Rey-Ares et al. (2021) show that self-control helps curb impulsive spending and undersaving. Liu et al. (2019) find that individuals with stronger self-control save more. Atkinson et al. (2013) demonstrate that loan repayment discipline fosters saving among Guatemalan micro-borrowers. Similarly, Achtziger et al. (2015) establish that low self-control correlates with undersaving, indebtedness, and compulsive buying.

2.3. Risk Aversion and Undersaving

In this study, risk aversion reflects an economic actor’s preference for reducing or avoiding risk. As Dohmen et al. (2017) note, risk attitudes should significantly influence economic decisions, such as savings and investment. Platteau et al. (2017) suggest that vulnerable, low-income individuals and the MINAEs they promote often exhibit higher risk aversion than higher-income earners. Theoretically, more risk-averse individuals may have a stronger propensity to save. In practice, however, the relationship between risk aversion and undersaving can be more complex due to various intervening factors.

The 2010 survey conducted by Enhancing Financial Innovation & Access [EFInA] (2010) indicates that Nigerians predominantly utilise microsaving accounts to address emergencies (thus functioning as microinsurance) and to meet day-to-day household needs (thereby serving as a consumption-smoothing mechanism). Consistent with this, the life-cycle theory of saving (Modigliani & Brumberg, 1954) posits that individuals save in order to stabilise consumption patterns over their lifetime, with the corollary implication that higher levels of risk aversion are associated with greater saving. This proposition has been further substantiated by Friedman and National Bureau of Economic Research (1957), Yaari (1965, 1969), Blume and Friend (1975), Hurd (1989), Benartzi and Thaler (1995), and Chetty et al. (2014).

2.4. Commitment Savings Devices and Undersaving

In this study, a commitment savings device (CSD) is defined as a mechanism designed to enhance the discipline of REAs in increasing microsavings or moderating consumption. The PMSS, introduced here as a CSD, set a standard threshold of 12% of REAs’ daily income and encouraged each REA to commit to a fixed daily saving (at or above this threshold) for a mandatory 60-day period, with funds locked until the end. Although the PMSS—specifically co-designed with the REAs—offers no matching incentives or special interest rates, it is premised on the hypothesis that sustaining 60 consecutive days of committed saving may reduce the tendency to undersave. Its effectiveness is further hypothesised to depend on individual motivation and context. These assumptions align with “time inconsistency” and “self-control” theories, as the PMSS creates a barrier to immediate consumption, while prioritising delayed gratification and fostering continuous saving toward long-term goals. As Mischel (2014) argues, self-control can be mastered; accordingly, this study investigates whether the PMSS could help REAs cultivate sustainable microsaving behaviour. As noted in Section 2.2, digital or mobile-enabled savings devices are now widespread, accessible to both low- and higher-income earners.

3. Data and Methodology of the Study

3.1. Research Design

This study adopts an experimental survey design inspired by N. Ashraf et al. (2006) but tailored to the REAs’ informal or semi-formal financial context in Nigeria. While N. Ashraf et al. (2006) tested monthly commitment savings devices (CSDs) over 12 months, this study introduced a daily savings commitment over 60 consecutive days, a design innovation reflecting the daily and irregular income cycles of REAs. The shorter horizon enhances feasibility, increases compliance, and touches on the debate on whether high-frequency commitments mitigate self-control constraints more effectively than longer, less frequent schemes. The study by Buchholtz et al. (2021) also informed our methodological design through its application of logistic regression in analysing saving behaviour, while Panda et al. (2014) similarly employed outcomes of experimental interventions as independent variables in logistic regression models.

3.2. Sampling Frame and Population

The sampling frame was derived from Nigeria Interbank Settlement System [NIBSS] (2021) data, which reported 167,000 active POS devices nationwide. As this dataset remained the most recent publicly available source when the survey commenced in 2022, it was adopted as the reference point. In the absence of a 2022 public registry of REAs, a conservative assumption was made that half of devices (83,500) belonged to REAs and half to non-REA users (merchants, hospitals, shops, etc.). This assumption was considered reasonable given the wide distribution of both REA and non-REA POS users across the country. To approximate Lagos’s share, the national figure was evenly divided by the 37 sub-national entities in Nigeria (including Abuja, the Federal Capital Territory), yielding an estimated 2256 REAs in Lagos. We acknowledge that this uniform distribution likely understates Lagos’s economic prominence, but it offered a pragmatic reference point.

3.3. Site Selection

From the 42 slums identified by the World Bank Group [WBG] (2014), 3 were randomly selected for all sessions of the REA Working Group but were later excluded from surveys to prevent sensitisation. The REA Working Group, comprising selected REAs, neobanks, and other willing stakeholders, was established to capture diverse insights in line with the “participatory development” or “participatory financial inclusion” approach (Mukherjee, 2004; Mansuri & Rao, 2013; Dror et al., 2014; Osifodunrin & Lopes, 2025b). Of the remaining 39, 21 slums were randomly selected as the survey base, balancing cost and representativeness. As REAs were unevenly distributed across these slums, 36–50 of them were randomly registered in each slum, producing a provisional frame of 1003 REAs.

3.4. Randomisation Procedure

Respondents for the cognitive and pilot surveys were randomly drawn without replacement. For the main surveys, 17–24 REAs were randomly selected from each of the 21 slums using computer-generated random numbers. Randomisation was stratified by slum to ensure proportional representation. Within slums, replacement for non-respondents followed the same randomisation process to preserve balance. Treatment and control assignment occurred at the individual level, with enumerators systematically blinded to allocation to minimise bias.

3.5. Attrition and Exclusion

Of the 1003 eligible REAs, 455 provided complete baseline-to-endline data. Attrition stemmed from refusals, relocation, or inability to sustain participation. An additional 23 REAs experiencing substantial income shocks and 11 defaulting during the treatment were excluded to prevent confounding. It should be noted that (before the study) the individual (micro)saving accounts of REAs are primarily transactional, with many respondents maintaining only paltry sums or the required minimum balance. Overall, attrition analysis revealed no significant differences in baseline demographics, income, or risk preferences between treatment and control groups, suggesting that attrition was largely random. Nonetheless, robustness checks (Section 3.8) tested sensitivity to these exclusions.

3.6. Interventions

Two interventions were implemented:

• Comprehensive Microsaving Enlightenment Programme (CMEP): Delivered to both treatment and control groups, the CMEP provided detailed financial literacy on microsaving, precautionary savings as insurance, micropension, and self-control strategies.
• Programmed Microsaving Scheme (PMSS): Only treatment REAs were required to commit to daily savings for 60 consecutive days through automatic standing orders. In this study, standing orders are prior instructions given by customers to their financial institutions to execute recurring transactions at a specified interval (daily, weekly, or otherwise). As noted earlier in Section 1, the initial benchmark of 18% of income—drawn from Nigeria’s pension law for workers in the formal sector—was revised downward to 12% following participatory deliberations with stakeholders. This adjustment struck a balance between policy alignment, optimal saving threshold for future financial security, and practical feasibility for informal or semi-formal, low-income earners. The revision is justifiable, considering that even formal-sector employees are mandated to save only a minimum of 8%, while their employers are required to contribute an additional minimum of 10% on their behalf (Federal Government of Nigeria [FGN], 2014; National Pension Commission [PENCOM], 2018). This decision is also reflective of the observations of Thaler and Benartzi (2004) and Bernheim (1994) that economic actors are often aware when they undersave and largely understand what their optimal saving thresholds should be.

3.7. Survey Instruments and Data Collection

Three structured main surveys—baseline, midline, and endline—were administered.

• Baseline: Collected demographics, average daily income, risk aversion, and self-control measures. Pre-treatment saving patterns were also recorded. For all pre-treatment savings data obtained during the baseline survey, validation was conducted using official (read-only) documentary evidence sourced by REAs from the digital platforms of neobanks and other microsaving providers. Wherever feasible, comparable validation procedures were applied to other categories of data collected at baseline, thereby enhancing the reliability and credibility of the dataset.
• Midline: Following repeated CMEP sessions, REAs reported intended daily saving commitments. For example, an REA averagely earning NGN 500 and willing to save 10% would commit NGN 50 daily, regardless of income fluctuations. Treatment REAs formalised these (with documentary evidence) via automatic standing orders with their PSPs, traditional banks, or neobanks.
• Endline: After 60 days of unbroken PMSS treatment, both groups reported post-treatment saving intentions as a fraction of average daily income.

To mitigate social desirability and other biases in self-control measures, enumerators used the cheap talk method and asked respondents to provide specific instances illustrating each survey question, reducing potential embarrassment.

3.8. Statistical Analysis

The primary dependent variable was a dichotomous indicator of undersaving (i.e., benchmark-defined undersaving):

• “1” if savings commitment was <12% of average daily income;
• “0” otherwise.

Logistic regression was the main analytical tool, with independent variables drawn from the baseline survey (income, demographics, risk aversion, and self-control). To strengthen validity, we performed the following:

• Robustness checks included probit models and OLS using continuous saving ratios.
• Alternative thresholds of 10% and 15% were used to test sensitivity around the 12% benchmark. Even the descriptive statistics indicate widespread benchmark-defined undersaving. At the 10% income-saving threshold, only 74 treated and 57 control REAs—131 in total, representing just 28.79% of the 455 successfully surveyed REAs—met or exceeded the benchmark. At the 12% threshold, the number fell to 58 treated and 50 control REAs, or 108 REAs (23.74%). At the 15% threshold, only 31 treated and 17 control REAs—48 in total, or 10.55%—attained or surpassed the benchmark.
• Clustered standard errors at the slum level addressed intra-cluster correlation.
• Attrition-bias tests confirmed that the results were robust to exclusions.
• Given the possible conceptual overlap between variables (average daily income, self-control, risk aversion, and others), we evaluated the potential for multicollinearity to affect our logistic regression estimates. Pairwise correlations among all independent variables were first examined, revealing modest associations (all |r| < 0.45). To quantify the extent of collinearity, Variance Inflation Factors (VIFs) were computed for each predictor; all VIF values were below 3, well within commonly accepted thresholds, indicating that multicollinearity is unlikely to materially inflate standard errors or bias coefficient estimates. Additionally, nested regressions excluding either self-control or risk aversion confirmed the stability of coefficient magnitudes and significance levels. These diagnostic checks support the robustness of our regression results and suggest that the observed statistical insignificance of certain predictors, such as risk aversion, reflects genuine behavioural patterns rather than artefacts of variable overlap.

Descriptive statistics (

Table 1. Income of REAs and frequency of undersaving.

Income of REAs and Frequency of Undersaving
S/N	Average Daily Income of REAs (NGN = NAIRA)	Frequency of REAs Willing to Undersave (or Willing to Save Below 12% of Their Average Daily Income) After the PMSS (i.e., POST_TS_DV_2 = 1)	Frequency of REAs Willing to Save 12% and Above of Their Average Daily Income After the PMSS (i.e., POST_TS_DV_2 = 0)	Total Frequency
1.	1000.00	9	7	16
2.	1250.00	4	1	5
3.	1500.00	22	6	28
4.	1550.00	1	0	1
5.	1600.00	1	0	1
6.	1700.00	1	0	1
7.	1750.00	18	4	22
8.	2000.00	17	0	17
9.	2250.00	14	4	18
10.	2500.00	33	15	48
11.	2600.00	1	0	1
12.	2750.00	8	2	10
13.	3000.00	26	8	34
14.	3500.00	45	10	55
15.	3700.00	1	0	1
16.	3750.00	2	0	2
17.	4000.00	59	20	79
18.	4500.00	39	11	50
19.	4750.00	17	4	21
20.	5000.00	12	6	18
21.	5500.00	12	6	18
22.	5750.00	2	1	3
23.	6000.00	3	2	5
24.	6500.00	0	1	1
	TOTAL	347	108	455

Note: USD 1.00 = ~NGN 414.72 as of 30 June 2022. POST_TS_DV_2 is the dichotomous dependent variable denoting the post-treatment saving decisions of REAs to undersave (coded as “1”) or the post-treatment saving decisions not to undersave (coded as “0”). Source: Authors’ computation from primary data collected during field study (2022/2023).

,

Table 2. PRCR_aversion and a proxy of post_treatment_savings.

PRCR_Aversion and a Proxy of Post_Treatment_Savings (Identifying Patterns in Measured Risk Aversion of REAs and Their Post-Treatment Savings Rate)
S/N	PRCR_Aversion (on a Likert Scale of 1 to 5)	Post_Treatment_Savings as a Percentage of REAs’ Daily Income (in Percentage Intervals)	Frequency of REAs in Each Interval
1.	1	0.00	2
		0.01–5.00	1
		5.01–11.99	10
		12.00–13.33	4
2.	2	0.00	2
		0.01–5.00	4
		5.01–11.99	10
		12.00–12.50	1
3.	3	0.00	7
		0.01–5.00	16
		5.01–11.99	57
		12.00–22.22	27
4.	4	0.00	6
		0.01–5.00	17
		5.01–11.99	69
		12.00–22.22	30
5.	5	0.00	20
		0.01–5.00	28
		5.01–11.99	98
		12.00–25.00	46
			TOTAL = 455

N.B: The frequency of REAs with measured PRCR_aversion of a Likert value of 1 is 17. The frequency of REAs with measured PRCR _aversion of a Likert value of 2 is 17. The frequency of REAs with measured PRCR_aversion of a Likert value of 3 is 107. The frequency of REAs with measured PRCR_aversion of a Likert value of 4 is 122. The frequency of REAs with measured PRCR_aversion of a Likert value of 5 is 192. Source: Authors’ computation from primary data collected during field study (2022/2023).

,

Table 3. Measured self-control and frequency of REAs by undersaving behaviour.

Measured Self-Control and Frequency of REAs by Undersaving Behaviour
S/N	Values of Self-Control as Measured by the Aggregation of 3 Sub-Items in the Survey on a Likert Scale of 1 to 5	Frequency of REAs Willing to Undersave (or Willing to Save Below 12% of Their Average Daily Income) After the PMSS (i.e., POST_TS_DV_2 = 1)	Frequency of REAs Willing to Save 12% and Above of Their Average Daily Income After the PMSS (i.e., POST_TS_DV_2 = 0)	Total Frequency
1.	1.00	6	0	6
2.	1.33	59	1	60
3.	1.67	75	2	77
4.	2.00	34	2	36
5.	2.33	86	0	86
6.	2.67	64	2	66
7.	3.00	7	11	18
8.	3.33	14	20	17
9.	3.67	0	14	14
10.	4.00	0	12	12
11.	4.33	1	16	17
12.	4.67	1	19	20
13.	5.00	0	9	9
	TOTAL	347	108	455

N.B: Self-control (denoted as AGGREGATE_SC) was measured in line with the instrument or sub-items used by Ameriks et al. (2007). POST_TS_DV_2 is the dichotomous dependent variable denoting the post-treatment saving decisions of REAs to undersave (coded as “1”) or the post-treatment saving decisions not to undersave (coded as “0”). Source: Authors’ computation from primary data collected during field study (2022/2023).

,

Table 4. Dynamics of undersaving behaviour for 455 successfully surveyed REAs.

Dynamics of Undersaving Behaviour for 455 Successfully Surveyed REAs
S/N	Data Characteristics	Frequency of REAs in the Treatment Group	Frequency of REAs in the Control Group	Total
1.	REAs that undersaved before the CMEP and the PMSS treatments	228	227	455
2.	REAs with reduced savings after the CMEP and the PMSS treatments (i.e., POST_TS < PRE_TS)	3	1	4
3.	REAs with the same savings even after the CMEP and the PMSS TREATMENTS (i.e., POST_TS = PRE_TS)	24	32	56
4.	REAs with increased savings after the CMEP and the PMSS but still undersaved (i.e., POST_TS > PRE_TS) and (12% of daily_income > POST_TS)	143 *	144	287
5.	REAs that did not undersave after the CMEP and the PMSS treatments (i.e., POST_TS > PRE_TS) and (12% of daily_income < = POST_TS)	58 *	50	108

PRE_TS = PRE_TS_DV_1 = pre-treatment savings of REAs. POST_TS = POST_TS_DV_2 = post-treatment savings of REAs. Note: All the 455 REAs received the CMEP treatment, but only 228 REAs received the PMSS treatment. * Future version(s) of this study will aim to monitor the progress made by these REAs to determine the long-term effect of the PMSS treatment on them. Source: Authors’ computation from primary data collected during field study (2022/2023).

,

Table 5. Demographic structure of 455 successfully surveyed REAs.

Demographic Structure of 455 Successfully Surveyed REAs
S/N	Demographic Variables	Descriptive Summaries	p-Values in the Regression Analysis
1.	Gender	279 or 61.32% were male and 176 or 38.68% were female.	0.420
2.	Religion	226 or 49.67% were Muslims and 229 or 50.32% were Christians.	0.675
3.	Marital status	104 or 22.86% were married and 351 or 77.14% were single.	0.512
4.	Formal education	- 16 or 3.52% had elementary school certificates or lower. - 223 or 49.01% had high school certificates or lower. - 34 or 7.47% had an Ordinary National Diploma (OND). - 182 or 40.00% had university degrees (or equivalent qualifications)	0.395
5.	Age	18 to 29 years = 263 REAs 30 to 39 years = 192 REAs 40 to 49 years = 0 REAs 50 to 59 years = 0 REAs above 59 years = 0 REAs	0.798

Note: p-value greater than 0.05 indicates statistical insignificance. Source: Authors’ computation from primary data collected during field study (2022/2023).

,

Table 6. REAs’ self-assessment for risk aversion.

REAs’ Self-Assessment for Risk Aversion
% of REA Respondents for Each Likert Point
S/N	Sub-Items of PRCP	Strongly Agree (5 Points)	Agree (4 Points)	Neutral (3 Points)	Disagree (1 Point)	Strongly Disagree (2 Points)	Total (455 REA Respondents)	Standard Deviation
1a	On a scale of 1 to 5 (1 being the least and 5 the highest), how would you assess your level of PRCR (risk) aversion?	42.19%	26.81%	23.52%	3.74%	3.74%	100.00%	1.07

Item 1a was to provide the REAs with an opportunity to self-assess their level of risk aversion (or PRCR aversion) as guided by the extant contributions of Dohmen et al. (2011) and Hardeweg et al. (2013), which affirmed that individuals are quite good at self-assessing their levels of risk aversion. Initially, anecdotal rationalisation might criticise this approach; however, the validity and reliability of the approach have been empirically affirmed by Bard and Barry (2000), Dohmen et al. (2011), and Falk et al. (2015). Details on measuring PRCR aversion (or risk aversion among REAs) were earlier presented by Osifodunrin and Lopes (2025a). Source: Authors’ computation from primary data collected during field study (2022/2023).

,

Table 7. Measured sub-items for risk (PRCR) aversion to validate the results in Table 6.

Measured Sub-Items for Risk (PRCR) Aversion to Validate the Results in Table 6
S/N	Sub-Items
2a	Do you always comply with all stipulations in the password policy of your payment service provider (PSP)? (This question is meant to show that the (non)compliance of an REA to this policy could be the difference between effectively avoiding PRCRs or not. The expected PRCR-free response is “yes”.)
2b	Do you have (in easy reach) all necessary information to successfully and promptly block or deactivate your REA account and related resources (if need be)? (This question rides on the fact that when the unexpected happens (whether in compromised REA accounts or even in situations of physically stolen POS devices, etc.), a PRCR-averse REA should have the ability to quickly disable all access to his/her account. The expected PRCR-free response is “yes”.)
2c	During personal emergencies, do you delegate your entire REA operations to a trusted third party? (As proper delegation requires the sharing of sensitive information (including passwords, etc.), possible compromise might occur, no matter the level of trust reposed in the appointed delegate. This question hints or exposes the possible PRCR that could materialise or crystallise from such action(s). The expected PRCR-free response is “no”.)
2d	Do you regularly document, review, and reconcile all financial transaction history emanating from your REA operations? (Ideally, any PRCR-averse REA should keep an independent personal record of all their transactions and reconcile this with relevant official records of transactions maintained on the PSP’s server and accessed by the REAs. This is to promptly identify any unauthorised or malicious transaction and prevent future occurrence of such. The expected PRCR-free response is “yes”.)
2e	Do you sometimes leave your POS device(s) unattended (This question arose because in the slums where the surveyed REAs operate, the rate of crime is rather high. In addition, some information inscribed on the POS device (such as the device manufacturer, year of manufacture, and the model/specification) is sensitive, and a potential hacker could use this information to investigate and review the list of (un)reported vulnerabilities available on the device to ultimately execute an informed attack. The expected PRCR-free response is “no”.)

In order to further affirm the validity and reliability of item 1a in Table 6 and to show that risk attitudes truly tend to be domain-specific (as opined and affirmed by Bard and Barry (2000) and Ding et al. (2010)), our study engaged sub-items 2a to 2e when (re)validating the results in Table 6. It is pertinent to note our decision, i.e., whenever the results from Table 6 align closely with the aggregated result in 2a to 2e, the REAs’ risk assessment in Table 6 is retained. Conversely, wherever we observed significantly diverse results (coming from Table 6 and then from 2a to 2e in Table 7), the REAs’ records in such instances were invalidated and extricated from the entire survey. Meanwhile, sub-items 2a to 2e are dichotomous (i.e., they require “yes” or “no” responses only). Details on measuring PRCR aversion (or risk aversion among REAs) were presented by Osifodunrin and Lopes (2025a). Source: Authors’ computation from primary data collected during field study (2022/2023).

and

Table 8. Measured sub-items for self-control.

Measured Sub-Items for Self-Control
		% of REA Respondents for Each Likert Point
S/N	Sub-Items	5 Points	4 Points	3 Points	2 Points	1 Point	Total (455 REA Respondents)	Standard Deviation
1	I do not often act without thinking through all the alternatives	6.16%	8.13%	16.48%	43.74%	25.49%	100.00%	1.11
2	I am good at resisting temptation	7.25%	17.80%	24.84%	31.21%	18.90%	100.00%	1.19
3	I am able to work diligently towards long-term goals	9.66%	14.29%	23.74%	34.73%	17.58%	100.00%	1.20

Sub-items 1 to 3 were as previously deployed and validated by Ameriks et al. (2007). Source: Authors’ computation from primary data collected during field study (2022/2023).

) complemented the regression outcomes provided in Table 9 and Table 10, while qualitative remarks from REAs contextualised findings.

3.9. Ethical Safeguards

This study adhered to ethical standards for social science research. Participants were assured anonymity, confidentiality, voluntary participation, and the right to withdraw without penalty. Informed consent was obtained verbally and recorded in questionnaires. The Research Ethics Committee of the University of Lisbon approved the study for ethical compliance and to fulfil the requirements for the research grant obtained from the Foundation for Science and Technology (Portugal).

3.10. Limitations

The methodology faces several limitations. First, population estimates relied on anecdotal assumptions about POS ownership and uniform distribution across the 37 sub-national entities in Nigeria. Second, the 60-day horizon captures short-term behaviour, limiting comparability with longer extant interventions. Third, dichotomising outcomes at a 12% saving threshold simplifies the analysis but sacrifices behavioural nuance. Fourth, while attrition analysis suggested no systematic bias, the exclusion of participants with extreme income shocks or defaults may understate real-world variability. Furthermore, participatory determination of the savings benchmark may have introduced social pressures affecting the reported commitments. Our literature-based insights remain constrained by the limited empirical research specifically examining the effects of income, demographic variables, self-control, risk aversion, and the experimental CSD intervention on the benchmark-defined undersaving of low-income groups. These limitations necessitate cautious interpretation of the results and encourage future replication.

3.11. Methodological Contributions

This study contributes methodologically by

• Introducing daily commitment contracts over 60 days—novel in informal-sector savings research.
• Combining top-down pension benchmarks with bottom-up participatory calibration or validation of savings thresholds.
• Embedding financial literacy (CMEP) alongside structural commitment devices (PMSS) and assessing their possible complementary effects using descriptive statistics.

These design features extend the experimental or methodological toolkit for studying microsaving in low- and irregular-income populations.

4. Results and Findings of the Study

The life-cycle hypothesis suggests that individuals save more when income is high and draw on those savings during retirement or income declines. Among the surveyed REAs, whose average daily income (Table 1) ranged from NGN 1000 to NGN 6500 (mean: NGN 3343.74 for 455 respondents), the baseline survey revealed pervasive undersaving. Of these, 280 REAs (61.54%) had zero savings, 114 (25.05%) saved less than 5% of daily income, and the remainder (i.e., 13.41% of 455) saved only 5.00–8.75%. Following the CMEP treatment—comprising intensive and targeted savings awareness for all REAs, together with a 60-day programme for all randomly selected and treated participants—the endline survey showed an improvement: 37 had zero savings, 54 saved less than 5%, 256 saved 5–11.76%, and 108 saved 12% or more. Despite these dynamics, logistic regression indicates that average daily income has no statistically significant effect on benchmark-defined undersaving (p = 0.634 > 0.05, Table 10). Likewise, Table 1 shows no clear pattern: for example, among 16 REAs earning NGN 1000 and 5 earning NGN 6000, the ratio of undersavers to non-undersavers was nearly identical despite the income gap. This supports time inconsistency and self-control theories, which hold that individuals across income levels face similar behavioural challenges. Without mechanisms such as CSD or other institutional and psychological supports, people may over-consume, over-borrow, or undersave while discounting the future. Banerjee and Duflo (2011) similarly note that saving decisions—both in frequency and magnitude—are difficult for unsupported low- and high-income individuals. Consistent with this, Kraay (2000) finds that future income growth has no significant effect on household saving rates, while Achtziger et al. (2015) show that income is not significantly linked to undersaving proxies such as compulsive buying or indebtedness. Furthermore, qualitative, scenario-based responses from several REAs suggest that irregular income flows, together with network and peer effects, influence—and in some cases moderate—the magnitude and frequency of their savings, notwithstanding the 60-day PMSS intervention. In addition, a few REAs noted that growing personal and family responsibilities may intensify as daily income rises, potentially offsetting gains in savings capacity.

For the demographic variables—age, marital status, formal education, gender, and religion—the logistic regression estimates indicate no statistically significant association with REAs’ benchmark-defined undersaving. Insights from the coded scenario-based qualitative responses complement this result, indicating that saving behaviour is shaped less by individual demographic attributes than by shared structural and occupational conditions. REAs consistently emphasised common income volatility, liquidity constraints, and immediate consumption pressures inherent in operating within the same informal, high-risk POS environment in urban slums. These constraints generate broadly similar cash-flow management challenges across agents, irrespective of age, family status, gender, educational attainment, or religious affiliation, thereby attenuating demographic differentials typically observed in formal or higher-income settings. In this context, undersaving appears more closely linked to short- and long-term cash-flow management capacity than to demographic characteristics per se, rendering the latter statistically insignificant in explaining benchmark-defined undersaving. This finding is consistent with Kong et al.’s (2023) findings, who likewise report no significant influence of demographic factors on customers’ saving or investment outcomes. An important qualification concerns age. While prior studies document significant age effects on saving behaviour consistent with life-cycle theory (Horioka & Watanabe, 1997; Ameriks et al., 2003), the absence of a statistically significant age effect in this study is plausibly attributable to the restricted age range of the surveyed REAs (18–39 years; Table 5). Given the predominance of younger agents in Lagos slums and the limited representation of older cohorts (40–60 years), the regression analysis lacks sufficient age variation to capture life-cycle saving dynamics, thereby constraining the detectability of age-related effects in this setting.

Rational choice frameworks suggest that economic agents respond to perceived risks by adopting precautionary strategies, including increased saving. This study examines POS-related cyber risks (PRCRs)—defined by Osifodunrin and Lopes (2023) as malicious and non-malicious cyber incidents inherent in REAs’ POS operations, such as data breaches, malware attacks, identity theft, and network downtime that may result in financial loss or business disruption. Contrary to theoretical expectations, the logistic regression results indicate that PRCR aversion, as measured in the survey, is not statistically significantly associated with REAs’ benchmark-defined undersaving. Evidence from the coded scenario-based qualitative responses helps to contextualise this result. REAs commonly (but not accurately) perceive POS-related risks as episodic, manageable, and largely preventable through vigilance, customer familiarity, and experience-based coping strategies, rather than as persistent threats warranting systematic precautionary savings. As a result, heightened awareness or aversion to PRCRs does not consistently translate into higher saving or reduced undersaving. Instead, immediate liquidity requirements and daily operating pressures tend to dominate forward-looking risk considerations, leading REAs to prioritise immediate pressures over building savings explicitly for (perceived) low-probability cyber or operational shocks. Consequently, even among risk-averse agents, saving behaviour remains largely unchanged, rendering measured PRCR aversion statistically insignificant in explaining benchmark-defined undersaving. This finding aligns with Kong et al.’s (2023) findings, who similarly report that survey-based measures of risk aversion are insignificantly correlated with investment or saving outcomes. Descriptively, Table 2 shows that as the Likert score of PRCR aversion increases from 1 to 5, both the “willingness not to undersave” and the “willingness to undersave” also increase, producing no monotonic pattern. In contrast, Horn (2022) documents a positive association between risk aversion and saving rates, suggesting that REAs’ responses to POS-related cyber risks may diverge from broader empirical findings. Nonetheless, the rapidly evolving nature of PRCRs implies that continued reliance on informal coping mechanisms may warrant reassessment as these risks become more frequent or severe.

According to the “time inconsistency” and “hyperbolic discounting” theories, REAs display a preference for immediate consumption, undervaluing future benefits from microsaving. While intending to save part of their daily income for pensions or target savings, they often struggle with self-control when confronted with immediate consumption temptations. Consequently, REAs—like many low-income individuals—undersave or fail to meet long-term financial goals (Laibson, 1996). As shown in Table 4, all REAs surveyed before the CMEP and PMSS treatments undersaved. Even after treatment, 347 (76.26%) of participants, including control group members exposed only to CMEP, still expressed willingness to undersave. The logistic regression results identify self-control as statistically significant in explaining benchmark-defined undersaving. As shown in Table 10, of all hypothesised variables, only self-control yielded a negative statistical significance: higher self-control reduces willingness to undersave or the benchmark-defined undersaving. Table 3 illustrates this clearly—low self-control values (1.0–2.67) corresponded with a greater willingness to undersave, while higher values (3.67–5.00) corresponded with lower undersaving. These findings align with Boto-García et al. (2022), who show that self-control significantly increases the likelihood of regular saving habits. However, they diverge from Steinert et al. (2022), who found no evidence that self-control serves as a transmission channel for higher savings.

Theoretically, the introduced Programmed Microsaving Scheme (PMSS)—a daily commitment savings device implemented over 60 consecutive days—was designed to strengthen REAs’ self-control, mitigate time inconsistency, increase saving participation and intensity, and ultimately reduce benchmark-defined undersaving. To operationalise this objective, a baseline survey revealed that all 455 successfully surveyed REAs saved below the 12% average daily income threshold, with 280 REAs (61.54%) reporting no savings whatsoever prior to the intervention.

Following the baseline survey, 250 REAs were randomly assigned to a control group (TREATMENT_CONTROL = 0), while the remainder were assigned to the PMSS treatment group (TREATMENT_CONTROL = 1), consistent with the experimental protocols outlined by Chow et al. (2017). The PMSS was implemented as a hard commitment device through a standing order issued by each treated REA to a licenced bank or neobank, rendering deposits operationally irreversible during the 60-day period.

Post-treatment evidence reveals a substantial improvement in saving participation and performance. As shown in Table 2, only 37 REAs (8.13%) reported zero savings at the endline, compared with 280 at the baseline. Furthermore, Table 4 indicates that 395 REAs (86.81%) increased their savings relative to the baseline, comprising 287 REAs who saved more but remained below the 12% threshold and 108 REAs who saved 12% or more of their average daily income. These descriptive results indicate that the intervention environment—comprising PMSS for the treatment group and CMEP for all participants—was effective in mobilising savings and reducing zero-saving behaviour.

However, the central focus of this study is not saving participation per se but benchmark-defined undersaving. Conceptually, undersaving is defined relative to a minimum savings threshold, below which individuals are classified as undersavers regardless of incremental improvements. When undersaving is operationalised as a binary outcome—saving at least 12% of average daily income versus not—the logistic regression results (Table 10) show that PMSS participation has no statistically significant effect on the likelihood of exiting undersaving. As shown in Table 4, although 143 treated REAs and 144 control REAs increased their savings, most remained below the benchmark threshold. Only 58 treated REAs and 50 control REAs crossed the 12% threshold at the endline, underscoring the limited marginal effect of PMSS on benchmark attainment.

This pattern suggests that while the PMSS strengthened saving discipline and intensity, its impact was insufficient to consistently propel participants beyond a relatively demanding savings benchmark considering the 60-day horizon. Several factors help rationalise this outcome.

First, despite the hard design of the PMSS and the low default rate (as indicated in Section 3), qualitative remarks indicate that some REAs initially committed to overly ambitious saving plans and subsequently adjusted contributions downward to more realistic levels, consistent with John (2020).

Second, the short duration of the intervention, combined with high-frequency daily deposits, may have supported increased participation without allowing sufficient time for deeper habit formation or sustained threshold attainment. Prior studies suggest that longer-duration commitment devices with a lower deposit frequency are more effective in generating durable saving outcomes (N. Ashraf et al., 2006; Burke et al., 2014). In addition, the literature on habit formation (even in saving behaviour) suggests that repetitive actions over extended periods reinforce automaticity and long-term behavioural change, but short-horizon interventions—especially those with a high frequency (e.g., daily saving)—may improve participation without reliably achieving deeper habit development or threshold attainment (Loibl et al., 2011; Verplanken & Orbell, 2003).

Third, REAs operate in a context characterised by low and highly volatile daily incomes (approximately NGN 1000–6500), which constrains their ability to sustain fixed-percentage saving targets. Income shortfalls and competing household demands frequently necessitate temporary reductions in deposits, resulting in incremental gains (e.g., movement from 0 to 5% to 8 to 10%) that nonetheless fall short of the 12% benchmark.

Finally, both treatment and control groups participated in the CMEP financial literacy programme, which likely enhanced savings awareness and motivation across the sample. The strong savings response observed among control group REAs—85.46% of whom increased savings—suggests that CMEP alone generated substantial behavioural change, thereby attenuating the PMSS’s incremental effect on benchmark-defined undersaving.

Taken together, the findings indicate that the PMSS was effective in increasing saving participation and intensity but statistically ineffective in reducing benchmark-defined undersaving within the study period. This divergence reflects not a failure of behavioural response but a misalignment between the intervention’s time horizon, the economic realities of REAs, and the rigidity of the undersaving threshold. Future research should therefore examine longer-duration interventions, alternative benchmark structures, and continuous outcome measures to better capture meaningful but partial improvements in savings behaviour.

Outline of SPSS’S Logistic Regression Results

In Table 9, the p-value of 0.00 (which is <0.05 as highlighted) indicated that the significant model of the survey can be further interpreted. In Table 10, AGGREGATE_SC recorded the only significant (negative) relationship with the REAs’ willingness to undersave (which is benchmark-defined undersaving), i.e., for every one-unit increase in AGGREGATE_SC, the REAs’ willingness to undersave (denoted by POST_TS_DV_2) decreased by a factor of 3.430. Although risk aversion (denoted by PRCR_AVERSION) had no statistically significant effect on the REAs’ willingness to undersave, it is shown in Table 10 that for every one-unit increase in PRCR_AVERSION, REAs’ willingness to undersave (denoted by POST_TS_DV_2) also increased by a factor of 1.117.

Table 9. Omnibus tests of model coefficients.

Omnibus Tests of Model Coefficients
		Chi-Square	df	Sig.
Step 1	Step	312.329	10	0.000
	Block	312.329	10	0.000
	Model	312.329	10	0.000

Note: This is a system table automatically generated from IBM SPSS Statistics Version 25.

Table 9. Omnibus tests of model coefficients.

Omnibus Tests of Model Coefficients
		Chi-Square	df	Sig.
Step 1	Step	312.329	10	0.000
	Block	312.329	10	0.000
	Model	312.329	10	0.000

Note: This is a system table automatically generated from IBM SPSS Statistics Version 25.

Table 10. Variables in the equation.

Variables in the Equation
Step 1	B	S.E.	Wald	df	Sig.	Exp (B)	Lower	Upper
Gender	0.327	0.406	0.650	1	0.420	1.387	0.626	3.073
Age	0.102	0.396	0.066	1	0.798	1.107	0.509	2.406
Religion	0.166	0.395	0.176	1	0.675	1.180	0.544	2.559
Marital	−0.299	0.455	0.430	1	0.512	0.742	0.304	1.811
Education	0.169	0.199	0.723	1	0.395	1.184	0.802	1.749
Income	0.000	0.000	0.226	1	0.634	1.000	1.000	1.000
PRCR_aversion	0.110	0.168	0.431	1	0.512	1.117	0.803	1.553
AGGREGATE_SC	−3.430	0.377	82.963	1	0.000	0.032	0.015	0.068
Pre-treatment savings	0.000	0.002	0.039	1	0.844	1.000	0.996	1.005
Treatment_control	−0.215	0.386	0.310	1	0.577	0.806	0.378	1.719
Constant	9.732	2.145	20.578	1	0.000	16846.750

Note: This is a system table automatically generated from IBM SPSS Statistics Version 25.

Table 10. Variables in the equation.

Variables in the Equation
Step 1	B	S.E.	Wald	df	Sig.	Exp (B)	Lower	Upper
Gender	0.327	0.406	0.650	1	0.420	1.387	0.626	3.073
Age	0.102	0.396	0.066	1	0.798	1.107	0.509	2.406
Religion	0.166	0.395	0.176	1	0.675	1.180	0.544	2.559
Marital	−0.299	0.455	0.430	1	0.512	0.742	0.304	1.811
Education	0.169	0.199	0.723	1	0.395	1.184	0.802	1.749
Income	0.000	0.000	0.226	1	0.634	1.000	1.000	1.000
PRCR_aversion	0.110	0.168	0.431	1	0.512	1.117	0.803	1.553
AGGREGATE_SC	−3.430	0.377	82.963	1	0.000	0.032	0.015	0.068
Pre-treatment savings	0.000	0.002	0.039	1	0.844	1.000	0.996	1.005
Treatment_control	−0.215	0.386	0.310	1	0.577	0.806	0.378	1.719
Constant	9.732	2.145	20.578	1	0.000	16846.750

Note: This is a system table automatically generated from IBM SPSS Statistics Version 25.

Lastly, the equation form of this logistic regression model is as follows:

POST_TS_DV_2 = 9.732 − 3.430 (AGGREGATE_SC) + ε. (where “ε.” is the random error variable earlier described by Koutsoyiannis (1977) and Wulandari (2022)).

5. Conclusions

Nigeria and many other sovereign members of the Alliance for Financial Inclusion (AFI) argue that widening access to microsaving products is a vital strategy for poverty alleviation and reducing socio-economic vulnerability. Yet, access alone is insufficient; account holders must also save optimally to secure their financial future. According to EFInA’s 2020 survey, only 32% of Nigerian adults subscribed to or saved with a regulated financial institution, and a significant share of this group undersaved (Enhancing Financial Innovation & Access [EFInA], 2020). In the pre-PMSS survey for this study, all 455 validly surveyed REAs undersaved relative to the policy-reflective group target of 12% of average daily income.

To address the gaps noted by Karlan et al. (2014) and Osifodunrin and Lopes (2022), this study examined three independent variables—risk aversion, self-control, and PMSS treatment—on REAs’ willingness to undersave or the benchmark-defined undersaving using pre- and post-PMSS surveys and logistic regression. The results showed income and demographic variables (age, gender, marital status, education, religion) to be statistically insignificant. In contrast, self-control had a statistically significant negative effect on benchmark-defined undersaving, while both PMSS treatment and risk aversion were insignificant.

A key limitation, consistent with Thaler and Benartzi (2004), is that the 12% savings threshold, though group-determined, may not reflect optimal rates for individual REAs. Saving targets must be periodically adjusted for changing realities—age, family size, emergencies, income, retirement goals, and assets. Accordingly, the endline survey confirmed no significant changes in REAs’ income or circumstances during the 60-day PMSS period. The finding that the PMSS had no significant effect may reflect the short treatment horizon, with savings required daily for 60 days. In contrast, N. Ashraf et al. (2006) tested a 12-month CSD with only 12 monthly deposits. Future research should assess the combined effects of treatment duration and savings frequency on CSD effectiveness. Further limitations include the non-simultaneous administration of baseline, treatment, and endline surveys, leaving REAs exposed to exogenous shocks during data collection. Moreover, while 79% of global adults now own accounts (Klapper et al., 2025), this study underscores that the core challenge has shifted from access to the problem of undersaving. Additionally, no separate control group was established for the CMEP since all surveyed REAs were exposed to the programme, which was deemed essential for their participation.

The findings of this study likely exhibit moderate external validity for REAs and similar low-income agents operating in urban or peri-urban slums across other developing countries, where daily or irregular income streams, high consumption pressures, and liquidity constraints closely mirror the Nigerian context. Like Nigerian REAs, such populations often face time-inconsistent preferences and limited self-control, suggesting that benchmark-defined undersaving and longer-term daily commitment savings devices may be broadly relevant. Additionally, the urban slum environment—characterised by dense population, relatively higher crimes, shared consumption pressures, and exposure to similar economic shocks—reinforces the applicability of these innovative, participatory, and behavioural insights to other low-income urban populations globally.

However, caution is warranted when generalising to rural or semi-rural populations, such as smallholder farmers who earn seasonal or lumpy incomes rather than daily cash flows. For these groups, savings behaviour may be influenced more strongly by income timing, harvest cycles, and storage costs, meaning that interventions like daily PMSS may be less feasible or effective, and alternative commitment structures (e.g., monthly or harvest-linked deposits) could be more appropriate. Nevertheless, the core insight—that self-control (a well-researched behavioural constraint) often dominates socio-demographic predictors of savings behaviour—remains broadly informative for designing financial inclusion strategies targeting low-income populations worldwide.

6. Sustainable Development Goals and Other Crucial Policy Implications for the Study

The PMSS treatment appears to have contributed to improved post-treatment savings participation and savings performance among REAs, even where benchmark-defined undersaving was not significantly moderated. Consequently, policymakers should recognise that improved savings performance/participation does more than merely strengthen the balance sheets of REAs or households. Within the framework of the Sustainable Development Goals (SDGs), and as underscored by Mohammed and Uraguchi (2018), Do (2023), and Stiglitz (2025), improved savings behaviour contributes meaningfully to poverty reduction (SDG 1), reduced inequality (SDG 10), and improved health resilience (SDG 3), particularly among vulnerable and low-income populations. Policymakers should further note that REAs can serve as effective conduits for diffusing commitment saving practices and SDG-related narratives, especially among other low-income groups that they serve.

The study examined two interventions: the comprehensive microsaving enlightenment programme (CMEP), delivered to all participants, and the 60-day PMSS, delivered only to the treatment group. Surprisingly, 144 REAs not treated with PMSS increased their microsaving, though still below the 12% threshold, and 50 control-group REAs (previously undersavers) reached at least 12% of their average daily income by the endline survey. These results support those of Babiarz and Robb (2014) and Steinert et al. (2018) and cautiously suggest that CMEP itself moderates undersaving behaviour. Policymakers, governments, financial service providers (FMSPs), development agencies, and other stakeholders should therefore prioritise microsaving education and awareness initiatives for low-income populations. Based on Table 4, CMEP and PMSS appear to have nearly equivalent effects on undersaving, highlighting the potential value of CMEP-comparable programmes and warranting further investigation into their impact.

Although PRCR aversion (PRCR_AVERSION) is statistically insignificant in the logistic regression analysis, closer inspection of Table 2 reveals meaningful descriptive heterogeneity that is consistent with, rather than contradictory to, the multivariate results. Among the 108 REAs (treated and control) who crossed the 12% savings benchmark, the majority exhibit higher levels of measured risk aversion: 46 REAs reported a Likert score of 5, while a further 30 reported a score of 4, implying that 70.37% of benchmark-crossing savers fall within the upper range of PRCR aversion. At the same time, increases in PRCR aversion from Likert scores of 1 to 5 are associated with simultaneous rises in both the “willingness not to undersave” and the “willingness to undersave”, producing no monotonic relationship overall. This pattern suggests that while high PRCR aversion may characterise REAs who are able to achieve higher saving levels, it is neither sufficient nor systematic enough to generate a statistically significant average effect in the multivariate model, where liquidity constraints and operating pressures dominate saving decisions. Nonetheless, consistent with behavioural and precautionary saving theories, these descriptive patterns indicate that policy interventions aimed at strengthening risk awareness, risk acculturation, experiential learning, or cognitive capacity may still support more optimal saving behaviour among REAs with sufficient cash-flow flexibility (Bommier et al., 2012; Bommier & Grand, 2019), even as survey-based measures of risk aversion remain insignificantly correlated with saving outcomes on average (Kong et al., 2023).

Bryan et al. (2010) pose an important question for future research: whether CSDs merely induce short-term behavioural changes or can foster long-term improvements and a sustainable reduction in undersaving. Consistent with this, the 60-day PMSS findings are mixed. While participation and saving amounts increased, most REAs did not surpass the benchmark, underscoring the importance of treatment duration, deposit frequency, and realistic saving targets. Nevertheless, future iterations of this study will aim to assess the long-term effects of PMSS and track the sustained progress of PMSS-treated REAs, particularly those who demonstrated measurable improvements in saving behaviour. Short-term commitment devices can increase saving participation/performance, but their effect on surpassing ambitious thresholds is limited in contexts of low and volatile incomes. Future research and policy design should therefore focus on longer-duration interventions, adaptive or individualised benchmarks, and integrated literacy programmes to foster sustained reductions in undersaving and promote future financial security among REAs and other low-income populations.