1. Introduction

1.1 Background and motivation

Payment technology is one of the few financial innovations that can reach people before traditional banking does. A phone-based wallet needs no branch, no collateral, and no credit history, which is why mobile money has been studied so intensively as a route into the formal financial system for populations that banks never served (Jack & Suri, 2014; Suri & Jack, 2016). Globally, account ownership and digital payment use have grown rapidly, and the gap between countries is now less about whether people have a phone than about what the financial system lets them do with it (Demirgüç-Kunt et al., 2022; Klapper et al., 2025).

Mongolia is an unusual place to ask this question. It is a landlocked and sparsely populated country: in 2009 almost half of its people lived in Ulaanbaatar and the rest were spread across a very large territory, which made branch banking an expensive way to reach them (Firpo, 2009). Many households still rely on nomadic or semi-nomadic herding, and weather shocks are part of life: in the 2024 Global Findex, about one in four Mongolian adults reported an extreme weather event in the previous three years, and about two-thirds of those affected lost income or property (Bessir et al., 2026).

What the latest data reveal, however, is that Mongolia is no longer a country where payments bring the unbanked into the financial system. Account ownership among adults rose from about 78% in 2011 to about 92% in 2014 and 93% in 2017, and has been about 98% since 2021 (Demirgüç-Kunt et al., 2018; Bessir et al., 2026). In 2024, 95% of adults made in-store digital merchant payments with a card or phone, more than half paid online for delivered goods, and over 70% paid utility bills directly from an account, against 39%, about a quarter, and 37% respectively across low- and middle-income economies (Bessir et al., 2026). Figure 1 shows the account series.

# ------------------------------------------------------------------
# REAL (published) anchor data -- transcribed from the cited sources.
# Values marked approx = TRUE are reported only as "98%" in the source.
# Please re-check against data.worldbank.org / FRED before final submission.
# ------------------------------------------------------------------
acc <- tribble(
  ~year, ~group,    ~value, ~approx, ~source,
  2011,  "All adults", 77.72, FALSE, "Findex 2017 series (Demirgüç-Kunt et al., 2018)",
  2014,  "All adults", 91.82, FALSE, "Findex 2017 series (Demirgüç-Kunt et al., 2018)",
  2017,  "All adults", 92.97, FALSE, "Findex 2017 series (Demirgüç-Kunt et al., 2018)",
  2021,  "All adults", 98.00, TRUE,  "Bessir et al. (2026): 98% since 2021",
  2024,  "All adults", 98.00, TRUE,  "Bessir et al. (2026): 98% since 2021",
  2011,  "Women",      82.37, FALSE, "Findex 2017 series (Demirgüç-Kunt et al., 2018)",
  2014,  "Women",      93.21, FALSE, "Findex 2017 series (Demirgüç-Kunt et al., 2018)",
  2017,  "Women",      94.95, FALSE, "Findex 2017 series (Demirgüç-Kunt et al., 2018)",
  2021,  "Women",      99.00, FALSE, "World Development Indicators (Findex)",
  2024,  "Women",      99.00, FALSE, "World Development Indicators (Findex)"
)

ggplot(acc, aes(year, value, colour = group, shape = approx)) +
  geom_line(linewidth = 1) + geom_point(size = 3) +
  scale_colour_manual(values = c("All adults" = pal[["navy"]], "Women" = pal[["orange"]]), name = NULL) +
  scale_shape_manual(values = c(`FALSE` = 16, `TRUE` = 1), labels = c("exact", "approx. (reported as 98%)"), name = NULL) +
  scale_x_continuous(breaks = c(2011, 2014, 2017, 2021, 2024)) +
  scale_y_continuous(limits = c(70, 100), labels = label_number(suffix = "%")) +
  labs(title = "Figure 1. Account ownership, Mongolian adults, 2011-2024 (published)",
       subtitle = "Share of adults with an account at a financial institution or mobile money provider",
       x = NULL, y = NULL,
       caption = "Sources: Global Findex via World Bank series; Bessir et al. (2026). 2021 and 2024 total reported as 98%.")

Two features of this history matter for the project. First, most of the jump in account ownership (2011 to 2014) came before the era of QR codes and bank apps, which suggests that other channels did the work: government payments of wages, pensions, and social transfers that go directly into bank accounts (Firpo, 2009; Bessir et al., 2026), and in particular the Child Money Programme, launched in 2005, which pays about MNT 100,000 per month to every child aged 0 to 17 into an account in the child’s name (Bessir et al., 2026). Second, the digital payment boom was layered on top of a population that already had accounts. The domestic sources cited in the literature describe its speed: cashless payments were still a minority of transactions in 2016 (44% versus 56% for cash), but by the end of 2023 there were reportedly 3.5 million internet banking customers and 1.9 million mobile banking customers, and roughly 90% of payments were made in non-cash forms (Delgerjargal, 2024). The Governor of the Bank of Mongolia reports that real-time transactions grew by 45% a year and card transactions by 24% over the previous five years, with 4.3 million payment cards in circulation, of which 2.2 million were actively used (Bank of Mongolia, 2024a). These are central-bank statements whose definitions differ from the Findex, so I treat them as context rather than as comparable series.

These figures describe access and adoption, not outcomes. They do not tell us whether digital payments have deepened people’s use of savings, credit, and insurance, or whether they have simply changed the channel through which the same cash moves.

1.2 Problem statement

How the problem was restructured. My earlier draft treated digital payments as a stepping stone into the formal system, following Agarwal and Assenova (2024), and expected account ownership to be an outcome. The evidence above shows that this margin has closed: with about 98% of adults holding an account, account ownership cannot be the outcome of interest after 2017. I therefore restructured the problem from entry (do digital payments bring people into the system?) to deepening (given almost universal access, do digital payments go together with more formal saving, borrowing, and resilience, and is that deepening healthy?). The literature review is reorganized accordingly.

The problem. Mongolia has moved almost completely to digital payments on top of near-universal account ownership, but it is not known whether this has deepened households’ formal saving, borrowing, and ability to absorb shocks, or merely changed how cash moves. The question is sharpest for the dispersed rural and nomadic population, for whom distance, rather than missing credit information, was the historical barrier to formal finance, and it carries a financial-stability twist: credit has grown very fast, and more credit is not automatically better for households (Section 1.4).

The evidence gap. Evidence on payment technology and inclusion is concentrated in Kenya and sub-Saharan Africa, or in multi-country panels. Mongolia appears in the second group, for example as one of 25 middle-income Asian economies in a study of payment technology and entrepreneurship (Chen & Tran, 2025), and as one of many countries in the global sample behind the stepping-stone evidence (Agarwal & Assenova, 2024). Pooling forces Mongolia’s estimated effect toward the regional average and treats everything specific to the country as noise. Searches for Mongolian digital finance also tend to return work on Inner Mongolia, China, a different country with a different regulator, currency, and banking system (for example Li et al., 2025; Borjigin et al., 2025), which makes the single-country literature look larger than it is. Work on Mongolia itself is thin, and what exists is largely about adoption intentions among urban users (Delgerjargal, 2024) or descriptive reports from international organizations (Bessir et al., 2026), rather than about inclusion outcomes.

1.3 Research questions

The project asks whether the stepping-stone logic from other settings survives in a country where access is already universal and where the historical constraint is geographic dispersion. Specifically:

RQ1 (deepening). Is greater use of digital payments associated with more formal saving, more formal borrowing, and stronger financial resilience, holding constant account ownership and demographics?
RQ2 (geography). Do these associations differ between Ulaanbaatar and the dispersed rural and nomadic population?
RQ3 (deepening or substitution). Is the pattern consistent with digital payments deepening financial engagement, or with a shallow substitution of one payment channel for another, and what kind of borrowing is involved (consumption or business)?

A descriptive question comes first and supports all three: how did account ownership, digital payment use, saving, and borrowing evolve across the Findex waves (2011 to 2024), and where does Mongolia stand against comparison countries?

1.4 Relevance for investment analysis

Mongolia is a small frontier market, and an investor with a five-year-plus horizon has to judge two things: how large and deep the market for retail financial services can become, and how unstable the economy around it is. Digital payments bear on the first. Recent policy documents show why the second cannot be ignored. Credit growth was 22.0% in 2023, 19.8% in 2024, and 17.0% in 2025, and was still about 17% in May 2026, with inflation at 11.2% (International Monetary Fund [IMF], 2026). The Bank of Mongolia raised its policy rate by 2 percentage points to 12% in March 2025 and reset the upper limit on the debt-service-to-income (DSTI) ratio for new consumer loans at 50%; in January 2026 it lowered that limit to 45% and raised reserve requirements, citing the risks that fast consumer lending poses for imports, inflation, and household financial vulnerability (Bank of Mongolia, 2025; 2026). The IMF has recommended aligning the DSTI limit for non-bank lenders with that of banks and bringing pension-backed loans into the limit to prevent excessive indebtedness among pensioners (IMF, 2025; 2026). Whether digital payments widen healthy participation in formal finance, or mainly ease access to consumer debt, therefore matters directly for the long-run size and quality of the banking and payments market.

This paper does not test equity returns and gives no investment advice; the investment relevance is motivational, and a natural later step would be to link inclusion outcomes to the growth and risk of banks and payment providers.

1.5 Approach, contribution, and structure

The study is a single-country analysis that combines Global Findex survey waves with Bank of Mongolia payment statistics and National Statistics Office data, supplemented by Mongolian-language regulatory sources that the English-language literature does not draw on. The contribution is twofold. Empirically, it gives Mongolia its own estimate rather than a regional average. Conceptually, it moves the stepping-stone question from entry to deepening in a country with near-universal access, using evidence on account use (Dupas et al., 2018; Bachas et al., 2021) alongside the mobile money literature.

Section 2 reviews the literature and derives three propositions. Section 3 defines the data sources and variables, builds a synthetic test data set that mirrors the structure of the Findex microdata, and runs a simple preliminary analysis to test the empirical pipeline. References follow.

2. Literature Review

The literature relevant to this project falls into five strands: (i) the theory of institutional voids and the role of platforms in filling them; (ii) micro evidence on mobile money; (iii) evidence on what happens after people have accounts, which is the margin that matters in Mongolia; (iv) cross-country and macro evidence on payment technology, institutions, and entrepreneurship; and (v) the limited work on Mongolia. The strands build on one another, and the gaps between them define the contribution of this study. Table 1 is a map of the main sources.

lit <- tribble(
  ~Strand, ~Source, ~`Setting and design`, ~`Main message`, ~`Limit for Mongolia`,
  "1. Voids and platforms", "Khanna & Palepu (1997)", "Conceptual, emerging markets", "Missing intermediaries (information, enforcement) explain why markets and firms look different", "Concept, not a Mongolian test",
  "1. Voids and platforms", "Agarwal & Assenova (2024)", "151,771 individuals, 78 countries; regulatory change allowing non-bank mobile money", "Mobile money expands access to formal credit: a stepping stone, not a substitute", "Pooled; assumes missing credit information is the void",
  "2. Micro mobile money", "Jack & Suri (2014)", "Kenya household panel, M-PESA", "Lower transaction costs improve risk sharing", "Single non-bank operator; large unbanked population",
  "2. Micro mobile money", "Suri & Jack (2016)", "Kenya, long-run panel", "Agent access raised consumption and lifted households out of poverty, mostly female-headed", "Population largely outside the formal system before M-PESA",
  "2. Micro mobile money", "Ahmad et al. (2020)", "Review, African experience", "Gains vary with context and policy", "Africa-focused",
  "3. After access", "Dupas et al. (2018)", "Field experiment in Uganda, Malawi, Chile", "Free basic accounts, used by few, had no discernible average effects", "Shows access alone is not enough",
  "3. After access", "Bachas et al. (2021)", "Mexico: debit cards for transfer recipients who already had accounts", "Cards cut transaction and monitoring costs, built trust, and raised savings", "Closest analogue to Mongolia: accounts exist, usage is the margin",
  "3. After access", "Suri et al. (2021)", "Kenya, digital loans (M-Shwari), discontinuity design", "Digital loans raised credit without crowding out other loans and improved resilience", "Not a panacea; Kenya's market is thin",
  "4. Macro and firms", "Beck et al. (2018)", "General equilibrium model calibrated to Kenyan firms", "Payments technology matters for firms via trade credit and theft risk", "Calibrated to Kenya",
  "4. Macro and firms", "Chen & Tran (2025)", "25 middle-income Asian economies incl. Mongolia", "Mobile payments and regulatory quality raise entrepreneurship; traditional inclusion lowers it", "Cross-country; averages away Mongolia",
  "5. Mongolia", "Delgerjargal (2024)", "Survey of 390 digital wallet users in Ulaanbaatar", "Optimism and innovativeness raise intention to use; insecurity and discomfort reduce it", "Urban users only; intention, not outcomes",
  "5. Mongolia", "Bessir et al. (2026); Firpo (2009)", "World Bank descriptive posts using Findex 2024 and a 2009 field visit", "Universal accounts, exceptional digital use, high resilience; government payments drove early inclusion", "Descriptive; no identification"
)
lit %>% kable(align = "l", caption = "Table 1. Map of the literature") %>%
  kable_styling(bootstrap_options = c("striped", "condensed"), full_width = TRUE, font_size = 12) %>%
  column_spec(1, bold = TRUE, width = "10em") %>%
  column_spec(2, width = "10em")

Table 1. Map of the literature
Strand	Source	Setting and design	Main message	Limit for Mongolia
Voids and platforms	Khanna & Palepu (1997)	Conceptual, emerging markets	Missing intermediaries (information, enforcement) explain why markets and firms look different	Concept, not a Mongolian test
Voids and platforms	Agarwal & Assenova (2024)	151,771 individuals, 78 countries; regulatory change allowing non-bank mobile money	Mobile money expands access to formal credit: a stepping stone, not a substitute	Pooled; assumes missing credit information is the void
Micro mobile money	Jack & Suri (2014)	Kenya household panel, M-PESA	Lower transaction costs improve risk sharing	Single non-bank operator; large unbanked population
Micro mobile money	Suri & Jack (2016)	Kenya, long-run panel	Agent access raised consumption and lifted households out of poverty, mostly female-headed	Population largely outside the formal system before M-PESA
Micro mobile money	Ahmad et al. (2020)	Review, African experience	Gains vary with context and policy	Africa-focused
After access	Dupas et al. (2018)	Field experiment in Uganda, Malawi, Chile	Free basic accounts, used by few, had no discernible average effects	Shows access alone is not enough
After access	Bachas et al. (2021)	Mexico: debit cards for transfer recipients who already had accounts	Cards cut transaction and monitoring costs, built trust, and raised savings	Closest analogue to Mongolia: accounts exist, usage is the margin
After access	Suri et al. (2021)	Kenya, digital loans (M-Shwari), discontinuity design	Digital loans raised credit without crowding out other loans and improved resilience	Not a panacea; Kenya’s market is thin
Macro and firms	Beck et al. (2018)	General equilibrium model calibrated to Kenyan firms	Payments technology matters for firms via trade credit and theft risk	Calibrated to Kenya
Macro and firms	Chen & Tran (2025)	25 middle-income Asian economies incl. Mongolia	Mobile payments and regulatory quality raise entrepreneurship; traditional inclusion lowers it	Cross-country; averages away Mongolia
Mongolia	Delgerjargal (2024)	Survey of 390 digital wallet users in Ulaanbaatar	Optimism and innovativeness raise intention to use; insecurity and discomfort reduce it	Urban users only; intention, not outcomes
Mongolia	Bessir et al. (2026); Firpo (2009)	World Bank descriptive posts using Findex 2024 and a 2009 field visit	Universal accounts, exceptional digital use, high resilience; government payments drove early inclusion	Descriptive; no identification

2.1 Institutional voids and digital platforms

The starting point is the idea of institutional voids. Khanna and Palepu (1997) argued that emerging markets often lack the intermediaries that make markets work in richer economies, such as reliable information providers and enforcement mechanisms, and that firms and business groups arise partly to compensate. For financial services, the relevant voids are credit bureaus, collateral registries, and enforceable contracts: without them, banks cannot assess unfamiliar borrowers and so do not lend to them.

Agarwal and Assenova (2024) extend this idea to digital platforms. Building on the logic of multisided platforms, whose value rises with participation on each side (Rochet & Tirole, 2003), they argue that mobile money platforms fill voids in three ways: by generating transaction data that certifies users who lack credit histories, by giving unified access to financial products offered by other providers, and by scaling through network effects to people previously excluded. Using survey data on 151,771 individuals in 78 countries and regulatory changes that allowed non-banks to operate mobile money, they find that mobile money expands users’ access to credit from formal institutions. In their account, mobile money is a stepping stone into the formal system, not a substitute for it.

This is the mechanism my project examines in Mongolia, with two qualifications. Its first channel, data-based certification, depends on missing credit information. A setting where the historical obstacle was distance rather than missing information, and where banks already hold rich transaction data on salaried account holders, may behave differently. And its notion of a “stepping stone” is about entry, which has already happened in Mongolia. That is why the project asks about deepening instead.

2.2 Micro evidence on mobile money

The most credible evidence that payment technology changes household outcomes comes from Kenya. Jack and Suri (2014) show that M-PESA lowers the transaction costs of sending money and thereby improves risk sharing: households with access to the service were better able to maintain consumption when income shocks hit. Suri and Jack (2016) follow households over a longer period and find that access to mobile money agents raised consumption and lifted a meaningful share of households out of poverty, with the largest effects among female-headed households, operating through savings and a shift of occupations toward business activity. Together these papers establish that the benefits of mobile money go beyond convenience, and they identify two channels, risk sharing and savings, that are measurable in survey data.

Two limits matter for Mongolia. First, the Kenyan results come from a setting in which a single non-bank operator dominated, whereas Mongolia’s payment landscape is organized around banks and bank-linked wallets and apps. Second, the evidence is about households that were largely outside the formal system before mobile money, while in Mongolia the large majority already had an account when the digital boom began. A review of the African experience reinforces the point that the gains from mobile money vary with context and policy (Ahmad et al., 2020), which is a reason to estimate the effect locally rather than import it.

2.3 After access: from accounts to use

When accounts are already universal, the relevant evidence is about what makes people use them. This strand is the one that fits Mongolia best, and my earlier draft did not include it.

Dupas et al. (2018) randomly offered basic bank accounts at no cost in Uganda, Malawi, and Chile. Few people used them intensively: over two years only 17%, 10%, and 3% of those offered an account made five or more deposits, and the authors find no discernible average effects on savings or downstream outcomes. Their conclusion is that policies focused merely on expanding access to basic accounts are unlikely to improve welfare noticeably. For Mongolia, the implication is that near-universal account ownership says little about welfare by itself.

Bachas et al. (2021) study the opposite case: debit cards given at scale to cash-transfer recipients in Mexico who already had a bank account. After two years with the card, beneficiaries had accumulated savings equal to about 2% of annual income, financed by lower current consumption. The authors identify two mechanisms: cards reduce the transaction costs of reaching money, and they reduce monitoring costs, so that people check balances often and come to trust the bank. This is close to the Mongolian situation, where wages, pensions, and the Child Money Programme arrive in accounts and cards and phones are used for most payments. It also gives a mechanism for P1 below that does not depend on missing credit information.

Credit is the margin where the evidence cautions against assuming that more is better. Suri et al. (2021) study the fully digital M-Shwari loan in Kenya and find that it did not crowd out other credit and that eligible households were 6.3 percentage points less likely to forgo expenses after negative shocks, but they conclude that digital loans are not a panacea for broader credit market failures. In Mongolia, where consumer credit has grown rapidly and regulators have tightened debt-service limits (Section 1.4), the quality of borrowing is part of the question.

2.4 Payment technology, institutions, and entrepreneurship

A further line of work connects payment technology to production and entrepreneurship, not only to household welfare. Beck et al. (2018) build a dynamic general equilibrium model in which mobile money dominates cash as a medium of exchange because it removes the risk of theft, and they show that more productive entrepreneurs with access to trade credit are the ones most likely to adopt it. Calibrated to Kenyan firm data, the model implies sizeable gains for entrepreneurial growth and macroeconomic development. The paper shows that payment technology can matter for firms through trade credit and enforcement, which links the household evidence in Sections 2.2 and 2.3 to the institutional argument in Section 2.1.

The closest regional evidence is Chen and Tran (2025), who study 25 middle-income Asian economies, Mongolia among them, and examine how payment technology, traditional financial inclusion, and institutions relate to new-business formation. Their results indicate that mobile payment systems and regulatory quality are positively associated with entrepreneurship, that traditional financial inclusion is negatively associated with it, and that regulatory quality interacts with payment technology. The implication, consistent with Agarwal and Assenova (2024), is that the effect of payment technology depends on the institutional environment instead of being uniform across countries. It also introduces a puzzle that a single-country study can help resolve: if traditional inclusion and digital inclusion pull in different directions at the cross-country level, which one is operating in Mongolia?

Both of these studies, and the global Findex data that underlie so much of the field (Demirgüç-Kunt et al., 2022; Klapper et al., 2025), are built to compare countries. That design is a strength for identifying general patterns but a weakness for any country whose circumstances differ sharply from the average, because within-country variation is averaged away.

2.5 Mongolia: what is known and what is missing

Work specific to Mongolia is scarce, and it comes from three kinds of source.

Academic. The most relevant study I found examines the factors behind intention to use digital wallets, using a survey of 390 users in Ulaanbaatar and the Technology Readiness Index together with the Technology Acceptance Model (Delgerjargal, 2024). It finds that optimism and innovativeness raise the intention to use digital wallets, while insecurity and discomfort reduce it. The contribution is useful for understanding adoption, and it also documents the recent expansion of non-cash payments. Its limits are equally informative: the sample is restricted to urban users who already use digital wallets, the outcome is behavioral intention rather than a financial inclusion outcome, and the author notes that rural and remote areas, where connectivity may be unreliable, are outside the data. Many other results for “digital financial inclusion in Mongolia” concern Inner Mongolia, China (for example Li et al., 2025; Borjigin et al., 2025) and should not be mistaken for evidence on Mongolia.

Practitioner and descriptive. Firpo (2009) observed that Mongolia was already unusually banked, with government payments to children and pensioners going straight into bank accounts, and that mobile phone penetration was lower than financial service penetration, with little interoperability between bank-owned payment networks. Industry reporting dates the first mobile banking services to 2007 to 2010 (Khan Bank, XacBank, Mongol Post, and the telecom operator MobiCom’s Mobifinance), with some products aimed explicitly at semi-nomadic herders, and notes that the Financial Regulatory Commission, rather than the central bank, led the opening to non-bank participants (Oxford Business Group, 2013). The most recent descriptive evidence is the World Bank’s analysis of the 2024 Findex, which documents near-universal accounts, exceptional digital use, formal borrowing about twice the average for low- and middle-income economies, formal saving broadly in line with other economies, and unusually strong financial resilience (Bessir et al., 2026).

Official. The Bank of Mongolia and the IMF document very fast credit growth, large consumer lending, and tightening macroprudential limits (Bank of Mongolia, 2025; 2026; IMF, 2025; 2026), but these sources concern aggregates and do not link payment behaviour to household outcomes.

No study I could find uses Findex-type microdata to ask whether digital payments are associated with deeper saving, borrowing, and resilience in Mongolia, or whether the association differs between the capital and the countryside. That absence is itself the central observation of this review.

2.6 Synthesis, research gap, and propositions

Institutional-void theory says that financial exclusion reflects missing market-supporting institutions (Khanna & Palepu, 1997). Platform research says that mobile money can substitute for some of those institutions by certifying users and connecting them to formal providers (Agarwal & Assenova, 2024). The Kenyan micro evidence shows that the household payoffs can be large (Jack & Suri, 2014; Suri & Jack, 2016). Evidence on accounts says that access alone is not enough and that use is shaped by transaction costs and trust (Dupas et al., 2018; Bachas et al., 2021), and the firm-level and Asian cross-country evidence suggests that payoffs depend on credit access and regulatory quality (Beck et al., 2018; Chen & Tran, 2025). What is missing is evidence for a country with near-universal accounts, a geographic void, and a fast-growing consumer credit market. This yields three propositions for the empirical work:

P1 (deepening). If digital payments deepen financial engagement in Mongolia, greater use of digital payments should be associated with more formal saving and formal borrowing and with greater reliance on savings in an emergency, after controlling for account ownership and demographics.
P2 (geography). If geographic dispersion was the binding constraint, the association should be stronger outside Ulaanbaatar, where branch banking was never viable, than in the capital.
P3 (substitution). If digital payments mainly substitute one channel for another, usage will be associated with little additional saving or borrowing. A further warning sign is an association with borrowing but not with saving or resilience, which would suggest that digital payments ease access to consumer debt rather than build financial strength.

3. Data and Preliminary Analysis

3.1 Overview of the empirical design

No single Mongolian dataset can answer all three research questions, so the study uses complementary designs. The first is an individual-level analysis of Global Findex microdata, which observes the outcomes the literature uses (saving, borrowing, payments, and emergency money) and allows a rural and urban comparison. The second is an aggregate regional analysis of Bank of Mongolia and National Statistics Office (NSO) data across provinces (aimags) and years, which can capture variation the Findex cannot. A third element is an event-style analysis around dated payment-system changes. Using several designs is a deliberate response to the weakness described in Section 2.4: cross-country designs average away within-country variation, so identification here must come from variation inside Mongolia.

The design is associational. Adopting digital payments is a choice, and adopters differ from non-adopters in ways that also affect saving and borrowing, so I do not claim to estimate a causal effect.

Desk research for this project established three facts that shape the design:

Microdata exist for the latest wave. The World Bank Microdata Library lists the 2024 Global Findex microdata for Mongolia (reference ID MNG_2024_FINDEX_v02_M), with about 1,000 respondents (World Bank, 2025a).
Account ownership is a ceiling variable. With about 98% of adults holding an account since 2021 (Figure 1), a regression with account ownership as the outcome has almost no variation to explain after 2017. Account ownership is therefore a control, and the outcomes are saving, borrowing, and resilience.
The treatment is nearly saturated in 2024. If about 95% of adults make digital merchant payments (Bessir et al., 2026), the binary “uses digital payments” indicator has few non-users in the latest wave. Identification comes mainly from earlier waves and from a depth measure (the number of different digital uses), not from a simple user versus non-user contrast.

3.2 Data sources

sources <- tribble(
  ~Source, ~`What it provides`, ~`Level and frequency`, ~Status,
  "Global Findex Database 2025, Mongolia microdata (survey year 2024)",
  "Account ownership (`account`, `account_fin`, `account_mob`, `dig_account`); saving (`saved`, `fin17a` and related); borrowing (`borrowed`, `fin22a`, `fin22a_1`, `fin22e`); card or phone use of an account (`fin3`); receipt of wages, transfers, pensions; main source of emergency money (`fin24`); rural indicator (`urbanicity`); income quintile (`inc_q`); workforce (`emp_in`); education",
  "Individuals; repeated cross-section; about 1,000 adults per wave",
  "Catalog entry and variable list verified (World Bank, 2025a). To confirm: coding of `urbanicity`, names of the digital merchant payment items, sampling weights",
  "Earlier Findex waves (2011, 2014, 2017, 2021)",
  "Same indicators; published Mongolia series exist for 2011, 2014, 2017 and 2021",
  "Individuals; repeated cross-sections",
  "Published series verified (Figure 1). To confirm: microdata availability and variable harmonization across waves",
  "Bank of Mongolia: payment system statistics, annual reports, financial stability and monetary policy statements",
  "Transactions by channel, cards and users, deposits and loans, banking sector indicators, policy decisions on DSTI limits and reserves",
  "National, monthly to annual; some enterprise data by soum, district and village from 2024 (Bank of Mongolia, 2024b)",
  "Reports and statements verified. To confirm: whether payment channel series exist by aimag; household-level regional data",
  "National Statistics Office of Mongolia (NSO)",
  "Population by aimag, urban and rural share, household income and poverty, herding households and livestock, mobile subscriptions, internet use",
  "Aimag and national; annual",
  "Planned; exact tables still to be inventoried (nso.mn)",
  "Financial Regulatory Commission and Bank of Mongolia (Mongolian-language regulations and announcements)",
  "Dates of new payment rails, interoperability rules, licensing of non-bank providers",
  "Policy events",
  "Planned; relies on reading Mongolian primary sources (Table 3)",
  "IMF Financial Access Survey (via FRED) and IMF Article IV reports",
  "Household loans from commercial banks (% of GDP); credit growth; macroprudential policy",
  "National, annual",
  "Verified and used in Figure 2"
)
sources %>% kable(caption = "Table 2. Defined data sources", align = "l") %>%
  kable_styling(bootstrap_options = c("striped", "condensed"), full_width = TRUE, font_size = 12) %>%
  column_spec(1, bold = TRUE, width = "11em")

Table 2. Defined data sources
Source	What it provides	Level and frequency	Status
Global Findex Database 2025, Mongolia microdata (survey year 2024)	Account ownership (`account`, `account_fin`, `account_mob`, `dig_account`); saving (`saved`, `fin17a` and related); borrowing (`borrowed`, `fin22a`, `fin22a_1`, `fin22e`); card or phone use of an account (`fin3`); receipt of wages, transfers, pensions; main source of emergency money (`fin24`); rural indicator (`urbanicity`); income quintile (`inc_q`); workforce (`emp_in`); education	Individuals; repeated cross-section; about 1,000 adults per wave	Catalog entry and variable list verified (World Bank, 2025a). To confirm: coding of `urbanicity`, names of the digital merchant payment items, sampling weights
Earlier Findex waves (2011, 2014, 2017, 2021)	Same indicators; published Mongolia series exist for 2011, 2014, 2017 and 2021	Individuals; repeated cross-sections	Published series verified (Figure 1). To confirm: microdata availability and variable harmonization across waves
Bank of Mongolia: payment system statistics, annual reports, financial stability and monetary policy statements	Transactions by channel, cards and users, deposits and loans, banking sector indicators, policy decisions on DSTI limits and reserves	National, monthly to annual; some enterprise data by soum, district and village from 2024 (Bank of Mongolia, 2024b)	Reports and statements verified. To confirm: whether payment channel series exist by aimag; household-level regional data
National Statistics Office of Mongolia (NSO)	Population by aimag, urban and rural share, household income and poverty, herding households and livestock, mobile subscriptions, internet use	Aimag and national; annual	Planned; exact tables still to be inventoried (nso.mn)
Financial Regulatory Commission and Bank of Mongolia (Mongolian-language regulations and announcements)	Dates of new payment rails, interoperability rules, licensing of non-bank providers	Policy events	Planned; relies on reading Mongolian primary sources (Table 3)
IMF Financial Access Survey (via FRED) and IMF Article IV reports	Household loans from commercial banks (% of GDP); credit growth; macroprudential policy	National, annual	Verified and used in Figure 2

3.3 Dated payment-system milestones for the event-style analysis

The event-style analysis needs reliable dates. Table 3 lists the milestones I could verify in English-language sources, and what must still be confirmed in Mongolian primary sources before any date is used as an “event”.

ms <- tribble(
  ~Year, ~Milestone, ~Source, ~`Next step`,
  "2005", "Child Money Programme starts paying every child 0-17 into an account", "Bessir et al. (2026)", "Confirm design changes and top-ups",
  "2007-2010", "First bank mobile services (Khan Bank 2007; XacBank and Mongol Post 2008; MobiCom Mobifinance 2010)", "Oxford Business Group (2013)", "Confirm launch dates in primary sources",
  "2009", "Bank of Mongolia card issuance rules (Resolution 440); real-time gross settlement system for high-value payments", "Oxford Business Group (2013)", "Confirm in Bank of Mongolia documents",
  "2019", "Golomt Bank introduces SocialPay QR scanning domestically", "gogo.mn (n.d.)", "Confirm; add QPay interoperable QR date",
  "2021", "Bank of Mongolia migrates its payment card system to a new platform", "FinTech Futures (2021)", "Confirm go-live date",
  "2024", "Trade and Development Bank of Mongolia is first to offer Apple Pay", "Finextra (2024)", "Confirm date and uptake"
)
ms %>% kable(caption = "Table 3. Payment-system milestones (verified in English-language sources; to be confirmed in Mongolian primary sources)", align = "l") %>%
  kable_styling(bootstrap_options = c("striped", "condensed"), full_width = TRUE, font_size = 12) %>%
  column_spec(1, width = "6em")

Table 3. Payment-system milestones (verified in English-language sources; to be confirmed in Mongolian primary sources)
Year	Milestone	Source	Next step
2005	Child Money Programme starts paying every child 0-17 into an account	Bessir et al. (2026)	Confirm design changes and top-ups
2007-2010	First bank mobile services (Khan Bank 2007; XacBank and Mongol Post 2008; MobiCom Mobifinance 2010)	Oxford Business Group (2013)	Confirm launch dates in primary sources
2009	Bank of Mongolia card issuance rules (Resolution 440); real-time gross settlement system for high-value payments	Oxford Business Group (2013)	Confirm in Bank of Mongolia documents
2019	Golomt Bank introduces SocialPay QR scanning domestically	gogo.mn (n.d.)	Confirm; add QPay interoperable QR date
2021	Bank of Mongolia migrates its payment card system to a new platform	FinTech Futures (2021)	Confirm go-live date
2024	Trade and Development Bank of Mongolia is first to offer Apple Pay	Finextra (2024)	Confirm date and uptake

3.4 Variables

Table 4 maps each concept to a variable name in the 2024 Findex microdata. Names marked with an asterisk appear in the catalog’s variable list; items I could not verify are left to be confirmed in the data dictionary.

vars <- tribble(
  ~Block, ~Concept, ~`Definition used`, ~`Findex 2024 variable`,
  "Treatment (binary)", "Any digital payment", "Made or received at least one digital payment in the past year", "Derived indicator; name to confirm",
  "Treatment (depth)", "Digital uses (0-3)", "Count of distinct digital uses: used a card or phone to access an account, made a digital merchant payment, paid a bill from an account", "`fin3`*, merchant and bill items (to confirm)",
  "Outcome 1", "Formal saving", "Saved at a bank or similar financial institution in the past year", "`fin17a` (and `saved`* for any saving)",
  "Outcome 2", "Formal borrowing", "Borrowed from a bank or similar formal institution in the past year", "`fin22a`*, `fin22a_1`*, `borrowed`*",
  "Outcome 3", "Resilience", "Savings are the main source of emergency money within 30 days", "`fin24`* (category 1)",
  "Outcome 4", "Borrowing purpose", "Borrowed to start or operate a business, or for health", "`fin22e`*, `fin22d`*",
  "Control", "Account ownership", "Has an account", "`account`*, `account_fin`*, `account_mob`*",
  "Control", "Government payments", "Received a government transfer or pension", "`receive_transfers`*, `receive_pensions`*, `receive_wages`*",
  "Heterogeneity", "Rural", "Respondent lives in a rural area", "`urbanicity`* (check coding)",
  "Controls", "Demographics", "Age, gender, education, income quintile, workforce", "`age`, `female`, education*, `inc_q`*, `emp_in`*"
)
vars %>% kable(caption = "Table 4. Variable definitions and Findex 2024 variable names (* = verified in the catalog variable list)", align = "l") %>%
  kable_styling(bootstrap_options = c("striped", "condensed"), full_width = TRUE, font_size = 12) %>%
  column_spec(1, bold = TRUE, width = "9em")

Table 4. Variable definitions and Findex 2024 variable names (* = verified in the catalog variable list)
Block	Concept	Definition used	Findex 2024 variable
Treatment (binary)	Any digital payment	Made or received at least one digital payment in the past year	Derived indicator; name to confirm
Treatment (depth)	Digital uses (0-3)	Count of distinct digital uses: used a card or phone to access an account, made a digital merchant payment, paid a bill from an account	`fin3`*, merchant and bill items (to confirm)
Outcome 1	Formal saving	Saved at a bank or similar financial institution in the past year	`fin17a` (and `saved`* for any saving)
Outcome 2	Formal borrowing	Borrowed from a bank or similar formal institution in the past year	`fin22a`, `fin22a_1`, `borrowed`*
Outcome 3	Resilience	Savings are the main source of emergency money within 30 days	`fin24`* (category 1)
Outcome 4	Borrowing purpose	Borrowed to start or operate a business, or for health	`fin22e`, `fin22d`
Control	Account ownership	Has an account	`account`, `account_fin`, `account_mob`*
Control	Government payments	Received a government transfer or pension	`receive_transfers`, `receive_pensions`, `receive_wages`*
Heterogeneity	Rural	Respondent lives in a rural area	`urbanicity`* (check coding)
Controls	Demographics	Age, gender, education, income quintile, workforce	`age`, `female`, education, `inc_q`, `emp_in`*

3.5 Published anchor data: household credit and macro context

Before any simulation, Figure 2 shows the published series on household borrowing and credit growth that motivate the “healthy deepening” question.

macro <- bind_rows(
  tribble(~year, ~value, ~panel, ~proj,
          2020, 22.78, "A. Bank loans to households (% of GDP)", FALSE,
          2021, 23.86, "A. Bank loans to households (% of GDP)", FALSE,
          2022, 21.46, "A. Bank loans to households (% of GDP)", FALSE,
          2023, 21.83, "A. Bank loans to households (% of GDP)", FALSE,
          2024, 27.46, "A. Bank loans to households (% of GDP)", FALSE),
  tribble(~year, ~value, ~panel, ~proj,
          2023, 22.0, "B. Credit growth (% change)", FALSE,
          2024, 19.8, "B. Credit growth (% change)", FALSE,
          2025, 17.0, "B. Credit growth (% change)", FALSE,
          2026, 16.4, "B. Credit growth (% change)", TRUE)
)
ggplot(macro, aes(factor(year), value, fill = proj)) +
  geom_col(width = 0.65) +
  geom_text(aes(label = number(value, accuracy = 0.1)), vjust = -0.4, size = 3.5) +
  facet_wrap(~panel, scales = "free") +
  scale_fill_manual(values = c(`FALSE` = pal[["navy"]], `TRUE` = pal[["light"]]), guide = "none") +
  scale_y_continuous(expand = expansion(mult = c(0, 0.15))) +
  labs(title = "Figure 2. Household credit and credit growth in Mongolia (published)",
       x = NULL, y = NULL,
       caption = "Sources: IMF Financial Access Survey via FRED (panel A); IMF staff concluding statement, June 2026 (panel B; 2026 is a projection, light bar).")

Bank loans to households rose from about 21.5% of GDP in 2022 to 27.5% in 2024, while overall credit growth, though easing, remained high. In the Findex, saving at a financial institution was about 19% in 2017 and 28% in 2021 (World Bank Global Financial Development series via FRED). These published margins are used below to calibrate the synthetic data.

3.6 The synthetic test data set

The real Findex microdata for Mongolia are not yet downloaded, so to test the whole workflow (cleaning, descriptive tables, regressions, figures) I build a synthetic data set with the structure of the Findex microdata.

set.seed(20261109)

waves <- c(2011, 2014, 2017, 2021, 2024)
n_w   <- 1000                                   # about 1,000 adults per Findex wave

# Margins by wave. "published" = taken from the sources above; "assumed" = arbitrary.
targets <- tibble(
  wave        = waves,
  p_account   = c(.777, .918, .930, .980, .980), # published (Figure 1)
  p_digital   = c(.04, .12, .30, .60, .93),      # assumed; 2024 echoes ~95% digital merchant payments
  p_save      = c(.15, .17, .193, .280, .300),   # 2017 and 2021 published; others assumed
  p_borrow    = c(.15, .20, .25, .32, .36),      # assumed
  p_emerg_sav = rep(.40, 5)                      # assumed
)

# Choose an intercept so that the mean predicted probability hits the target margin
calib <- function(lp, target) {
  uniroot(function(a) mean(plogis(a + lp)) - target, c(-20, 20))$root
}

sim_wave <- function(w) {
  tg    <- filter(targets, wave == w)
  w_idx <- match(w, waves) - 1

  d <- tibble(
    wave   = w,
    rural  = rbinom(n_w, 1, 0.32),
    female = rbinom(n_w, 1, 0.52),
    age    = round(pmin(pmax(rnorm(n_w, 41, 15), 18), 80)),
    educ   = sample(1:3, n_w, TRUE, prob = c(.15, .55, .30)),  # 1 primary or less, 2 secondary, 3 tertiary
    inc_q  = sample(1:5, n_w, TRUE),
    emp    = rbinom(n_w, 1, 0.55)
  )
  d$gov_transfer <- rbinom(n_w, 1, plogis(-0.2 + 0.9 * (d$age >= 60) + 0.3 * d$rural))

  # Treatment: any digital payment, then depth (number of distinct digital uses)
  lp_dig <- with(d, 0.25*(inc_q-3) + 0.5*(educ-2) - 0.6*rural - 0.03*(age-41) + 0.3*emp)
  d$digital_payment <- rbinom(n_w, 1, plogis(calib(lp_dig, tg$p_digital) + lp_dig))
  d$digital_uses    <- d$digital_payment *
    (1 + rbinom(n_w, 2, plogis(-1 + 0.35*w_idx - 0.3*d$rural + 0.2*(d$educ-2))))

  # Account ownership: near ceiling in later waves
  lp_acc <- with(d, 0.6*digital_payment + 0.25*(inc_q-3) + 0.3*(educ-2) - 0.3*rural + 0.3*gov_transfer)
  d$account <- rbinom(n_w, 1, plogis(calib(lp_acc, tg$p_account) + lp_acc))

  # Outcomes: effects of digital depth are ASSUMED (larger for saving, small for borrowing, larger in rural areas)
  lp_sav <- with(d, 0.25*digital_uses + 0.30*digital_uses*rural + 0.25*(inc_q-3) + 0.25*(educ-2) + 0.5*account + 0.1*emp)
  d$formal_saving <- rbinom(n_w, 1, plogis(calib(lp_sav, tg$p_save) + lp_sav))

  lp_bor <- with(d, 0.12*digital_uses + 0.10*digital_uses*rural + 0.15*(inc_q-3) + 0.2*emp + 0.3*account)
  d$formal_borrowing <- rbinom(n_w, 1, plogis(calib(lp_bor, tg$p_borrow) + lp_bor))
  d$borrow_business  <- d$formal_borrowing * rbinom(n_w, 1, plogis(-2 + 0.3*d$emp + 0.4*d$rural))

  lp_em <- with(d, 0.6*formal_saving + 0.15*digital_uses + 0.2*(inc_q-3))
  d$emergency_savings <- rbinom(n_w, 1, plogis(calib(lp_em, tg$p_emerg_sav) + lp_em))
  d
}

syn <- bind_rows(lapply(waves, sim_wave)) %>%
  mutate(id = row_number(), .before = 1)

glimpse(syn)

## Rows: 5,000
## Columns: 16
## $ id                <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,…
## $ wave              <dbl> 2011, 2011, 2011, 2011, 2011, 2011, 2011, 2011, 20…
## $ rural             <int> 0, 1, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 0, 1, 0,…
## $ female            <int> 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 1, 1,…
## $ age               <dbl> 18, 34, 47, 49, 52, 18, 41, 36, 61, 36, 20, 43, 55…
## $ educ              <int> 2, 2, 2, 2, 3, 3, 3, 2, 2, 2, 2, 3, 2, 3, 2, 2, 1,…
## $ inc_q             <int> 4, 5, 2, 2, 1, 4, 1, 5, 2, 1, 2, 1, 4, 5, 1, 3, 1,…
## $ emp               <int> 1, 0, 0, 0, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1,…
## $ gov_transfer      <int> 1, 0, 1, 1, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0,…
## $ digital_payment   <int> 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1,…
## $ digital_uses      <dbl> 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 1,…
## $ account           <int> 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1,…
## $ formal_saving     <int> 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0,…
## $ formal_borrowing  <int> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ borrow_business   <int> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ emergency_savings <int> 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1, 1,…

Table 5. Synthetic data dictionary
Variable	Meaning
wave	Survey year (2011, 2014, 2017, 2021, 2024); 1,000 simulated adults each
digital_payment	1 = made or received a digital payment
digital_uses	Number of distinct digital uses, 0 to 3
account	1 = has an account
formal_saving, formal_borrowing	1 = saved or borrowed formally in the past year
borrow_business	1 = formal borrower who borrowed to start or run a business
emergency_savings	1 = savings are the main source of emergency money
rural, female, age, educ, inc_q, emp	Demographics (educ: 1 to 3; inc_q: income quintile; emp: in workforce)
gov_transfer	1 = received a government transfer or pension

3.7 Preliminary analysis on the synthetic data

3.7.1 Descriptive statistics and the two ceilings

desc <- syn %>%
  group_by(Wave = wave) %>%
  summarise(`Digital payment` = mean(digital_payment) * 100,
            `Account` = mean(account) * 100,
            `Formal saving` = mean(formal_saving) * 100,
            `Formal borrowing` = mean(formal_borrowing) * 100,
            `Savings = main emergency source` = mean(emergency_savings) * 100,
            `Mean digital uses` = mean(digital_uses),
            .groups = "drop")
desc %>% nice_table(caption = "Table 6. Synthetic data: share of respondents (%) by wave", digits = 1)

Table 6. Synthetic data: share of respondents (%) by wave
Wave	Digital payment	Account	Formal saving	Formal borrowing	Savings = main emergency source	Mean digital uses
2011	4.7	78.6	15.5	14.1	40.2	0.1
2014	10.7	91.2	15.6	19.0	40.9	0.2
2017	30.8	92.3	18.3	24.3	40.0	0.6
2021	57.7	97.9	28.1	33.7	41.1	1.2
2024	92.8	97.6	29.5	38.6	38.5	2.0

Two features of the real data show up here by construction. The account ceiling: by 2021 to 2024 almost everyone has an account, so account ownership cannot serve as an outcome. Treatment saturation: in 2024 only about 7% of the simulated sample does not use digital payments, so the binary treatment is identified mainly from earlier waves. This is why the main models use the depth measure and pool the waves with wave fixed effects.

long <- syn %>%
  group_by(wave) %>%
  summarise(across(c(digital_payment, account, formal_saving, formal_borrowing), mean), .groups = "drop") %>%
  pivot_longer(-wave, names_to = "series", values_to = "share") %>%
  mutate(series = recode(series, digital_payment = "Digital payment (assumed)",
                         account = "Account (calibrated)", formal_saving = "Formal saving (calibrated)",
                         formal_borrowing = "Formal borrowing (assumed)"))

published <- tribble(~wave, ~series, ~share,
                     2011, "Account (calibrated)", .7772, 2014, "Account (calibrated)", .9182,
                     2017, "Account (calibrated)", .9297, 2021, "Account (calibrated)", .98,
                     2024, "Account (calibrated)", .98,
                     2017, "Formal saving (calibrated)", .1934, 2021, "Formal saving (calibrated)", .2804)

ggplot(long, aes(wave, share, colour = series)) +
  geom_line(linewidth = 1) + geom_point(size = 2.5) +
  geom_point(data = published, shape = 1, size = 4.5, stroke = 1.1, colour = "black") +
  scale_x_continuous(breaks = waves) +
  scale_y_continuous(labels = percent_format(accuracy = 1), limits = c(0, 1)) +
  scale_colour_manual(values = unname(pal[c("navy", "orange", "green", "grey")]), name = NULL) +
  labs(title = "Figure 3. Synthetic data by wave (circled = published values used to calibrate)",
       x = NULL, y = NULL)

3.7.2 Model A: depth of digital payment use and formal outcomes

For respondent \(i\) in wave \(t\), I estimate linear probability models

\[Y_{it} = \alpha + \beta \cdot DigitalUses_{it} + X_{it}'\gamma + \delta_t + \varepsilon_{it},\]

where \(Y\) is formal saving, formal borrowing, or savings as the main emergency source, \(X\) includes age, gender, education, income quintile, workforce status, rural, account ownership and government payments, and \(\delta_t\) are wave fixed effects. Standard errors are heteroskedasticity-robust (HC1). A second specification uses the binary indicator \(DigPay_{it}\) instead of the depth measure.

stars <- function(p) ifelse(p < .01, "***", ifelse(p < .05, "**", ifelse(p < .10, "*", "")))
rob   <- function(m) coeftest(m, vcov = vcovHC(m, type = "HC1"))

est <- function(m, term, vc = vcovHC(m, type = "HC1")) {
  ct <- coeftest(m, vcov = vc)
  tibble(term = term, est = ct[term, 1], se = ct[term, 2], p = ct[term, 4],
         lo = est - 1.96 * se, hi = est + 1.96 * se)
}

ctrl_core <- "age + I(age^2/100) + factor(educ) + factor(inc_q) + female + emp"
ctrl_full <- paste(ctrl_core, "+ rural + account + gov_transfer + factor(wave)")
ctrl_B    <- paste(ctrl_core, "+ account + gov_transfer + factor(wave)")

outcomes <- c(`Formal saving` = "formal_saving",
              `Formal borrowing` = "formal_borrowing",
              `Savings = main emergency source` = "emergency_savings")

fit <- function(y, rhs) lm(as.formula(paste(y, "~", rhs)), data = syn)

mA_bin   <- lapply(outcomes, function(y) fit(y, paste("digital_payment +", ctrl_full)))
mA_depth <- lapply(outcomes, function(y) fit(y, paste("digital_uses +", ctrl_full)))
mB       <- lapply(outcomes, function(y) fit(y, paste("digital_uses * rural +", ctrl_B)))

cell <- function(e) c(paste0(sprintf("%.3f", e$est), stars(e$p)), paste0("(", sprintf("%.3f", e$se), ")"))

rows <- list(
  list("Any digital payment (binary)",           lapply(mA_bin,   function(m) est(m, "digital_payment"))),
  list("Digital uses, 0-3 (depth)",              lapply(mA_depth, function(m) est(m, "digital_uses"))),
  list("Model B: digital uses, urban (beta)",    lapply(mB,       function(m) est(m, "digital_uses"))),
  list("Model B: digital uses x rural (theta)",  lapply(mB,       function(m) est(m, "digital_uses:rural")))
)

tab <- do.call(rbind, lapply(rows, function(r) {
  cells <- lapply(r[[2]], cell)
  rbind(c(r[[1]], sapply(cells, `[`, 1)), c("", sapply(cells, `[`, 2)))
}))
colnames(tab) <- c("", names(outcomes))
tab <- rbind(tab, c("Observations", rep(format(nrow(syn), big.mark = ","), 3)))
tab %>% as.data.frame() %>%
  kable(caption = "Table 7. Linear probability estimates on synthetic data (robust standard errors in parentheses)", align = "lccc") %>%
  kable_styling(bootstrap_options = c("striped", "condensed"), full_width = FALSE, font_size = 13) %>%
  footnote(general = "* p<0.10, ** p<0.05, *** p<0.01. All models include demographics, account ownership, government payments and wave fixed effects. Synthetic data: the signs and sizes reflect assumptions built into the simulation.",
           general_title = "Notes: ", footnote_as_chunk = TRUE)

Table 7. Linear probability estimates on synthetic data (robust standard errors in parentheses)
V1	Formal saving	Formal borrowing	Savings = main emergency source
Any digital payment (binary)	0.139***	0.030*	0.063***
	(0.017)	(0.018)	(0.019)
Digital uses, 0-3 (depth)	0.069***	0.027***	0.035***
	(0.008)	(0.008)	(0.009)
Model B: digital uses, urban (beta)	0.051***	0.019**	0.033***
	(0.008)	(0.009)	(0.009)
Model B: digital uses x rural (theta)	0.065***	0.030**	0.009
	(0.013)	(0.013)	(0.014)
Observations	5,000	5,000	5,000
Notes: * p<0.10, p<0.05, * p<0.01. All models include demographics, account ownership, government payments and wave fixed effects. Synthetic data: the signs and sizes reflect assumptions built into the simulation.

The estimator recovers the pattern that was built in: depth of digital use is positively associated with formal saving and resilience, and only weakly with borrowing. In the real data, a pattern of borrowing without saving or resilience would be the warning sign in P3.

3.7.3 Model B: rural and urban differences (RQ2, P2)

Model B adds an interaction between digital uses and the rural indicator, \(\theta\). The figure shows the implied association separately for urban and rural adults, with 95% confidence intervals (the rural effect is \(\beta + \theta\), with its standard error from the variance of both coefficients).

lin_rural <- function(m) {
  V <- vcovHC(m, type = "HC1")
  b <- coef(m)
  t1 <- "digital_uses"; t2 <- "digital_uses:rural"
  e <- b[t1] + b[t2]
  s <- sqrt(V[t1, t1] + V[t2, t2] + 2 * V[t1, t2])
  tibble(est = unname(e), lo = unname(e - 1.96 * s), hi = unname(e + 1.96 * s))
}
coefplot <- bind_rows(
  bind_rows(lapply(mB, function(m) est(m, "digital_uses") %>% select(est, lo, hi)), .id = "outcome") %>% mutate(group = "Urban"),
  bind_rows(lapply(mB, lin_rural), .id = "outcome") %>% mutate(group = "Rural")
) %>% mutate(outcome = factor(outcome, levels = rev(names(outcomes))))

ggplot(coefplot, aes(est, outcome, colour = group)) +
  geom_vline(xintercept = 0, linetype = 2, colour = "grey50") +
  geom_pointrange(aes(xmin = lo, xmax = hi), position = position_dodge(width = 0.5), size = 0.7) +
  scale_colour_manual(values = c(Urban = pal[["navy"]], Rural = pal[["orange"]]), name = NULL) +
  labs(title = "Figure 4. One more digital use and each outcome (synthetic data)",
       x = "Change in probability (linear probability model)", y = NULL)

3.7.4 Robustness: logit marginal effects and coefficient stability

The sample is small, so I check that a logit gives similar average marginal effects, and I report how the depth coefficient moves as controls are added. The last column is an approximate version of the proportional-selection bound of Oster (2019): the value of \(\delta\) says how strong selection on unobservables would have to be, relative to selection on the observed controls, to push the coefficient to zero (with \(R_{max} = 1.3 \times\) the full-model \(R^2\)). Values above 1 are conventionally read as reassuring. The exact procedure is available in Stata (psacalc) and in the R package robomit.

ame_depth <- function(y) {
  f <- as.formula(paste(y, "~ digital_uses +", ctrl_full))
  m <- glm(f, data = syn, family = binomial)
  mean(predict(m, transform(syn, digital_uses = digital_uses + 1), type = "response") -
       predict(m, syn, type = "response"))
}

oster_delta <- function(b0, r0, b1, r1, rmax_mult = 1.3) {
  rmax <- min(1, rmax_mult * r1)
  if (b0 <= b1) return(NA_real_)           # controls did not reduce the coefficient
  b1 * (r1 - r0) / ((b0 - b1) * (rmax - r1))
}

stab <- lapply(names(outcomes), function(nm) {
  y  <- outcomes[[nm]]
  m0 <- fit(y, "digital_uses + factor(wave)")
  m1 <- fit(y, paste("digital_uses +", ctrl_core, "+ rural + factor(wave)"))
  m2 <- fit(y, paste("digital_uses +", ctrl_full))
  tibble(Outcome = nm,
         `Wave FE only` = coef(m0)["digital_uses"],
         `+ demographics` = coef(m1)["digital_uses"],
         `+ account, gov. payments` = coef(m2)["digital_uses"],
         `Logit AME` = ame_depth(y),
         `Approx. Oster delta` = oster_delta(coef(m0)["digital_uses"], summary(m0)$r.squared,
                                              coef(m2)["digital_uses"], summary(m2)$r.squared))
}) %>% bind_rows()

stab %>% nice_table(caption = "Table 8. Depth coefficient across specifications (synthetic data)", digits = 3)

Table 8. Depth coefficient across specifications (synthetic data)
Outcome	Wave FE only	demographics	account, gov. payments	Logit AME	Approx. Oster delta
Formal saving	0.077	0.069	0.069	0.063	8.727
Formal borrowing	0.033	0.027	0.027	0.024	3.994
Savings = main emergency source	0.045	0.035	0.035	0.035	10.177

3.7.5 Model C: a regional panel (test of the code only)

Where aimag-level data exist, I will estimate

\[Y_{pt} = \alpha_p + \delta_t + \beta \cdot D_{pt} + \theta \cdot (D_{pt} \times Herding_p) + \varepsilon_{pt},\]

for aimag \(p\) and year \(t\), with aimag and year fixed effects, standard errors clustered by aimag, and results reported with and without Ulaanbaatar. The block below tests the code on a synthetic panel of Ulaanbaatar plus 21 invented aimags observed for ten years. Aimag names are placeholders, not real provinces.

set.seed(777)
aimags <- tibble(aimag   = c("Ulaanbaatar", sprintf("Aimag %02d", 1:21)),
                 herding = c(0.01, runif(21, 0.10, 0.60)),
                 a_p     = rnorm(22, 1.2, 0.3),
                 g_p     = c(0.90, runif(21, 0.25, 0.75)))   # speed of digital adoption

panel <- expand_grid(aimags, year = 2015:2024) %>%
  mutate(yr = year - 2015,
         D = 0.10 + g_p * yr + rnorm(n(), 0, 0.05),                     # digital payment intensity (index)
         y = a_p + 0.03 * yr + 0.10 * D + 0.15 * D * herding + rnorm(n(), 0, 0.05))  # accounts per capita (assumed effects)

run_panel <- function(dat) {
  m <- lm(y ~ D + I(D * herding) + factor(aimag) + factor(year), data = dat)
  vc <- vcovCL(m, cluster = dat$aimag, type = "HC1")
  bind_rows(est(m, "D", vc), est(m, "I(D * herding)", vc))
}

bind_rows(
  run_panel(panel) %>% mutate(Sample = "All 22 units"),
  run_panel(filter(panel, aimag != "Ulaanbaatar")) %>% mutate(Sample = "Without Ulaanbaatar")
) %>%
  mutate(Term = recode(term, D = "Digital intensity (beta)", `I(D * herding)` = "Digital intensity x herding share (theta)"),
         Estimate = sprintf("%.3f%s", est, stars(p)), `Std. error` = sprintf("(%.3f)", se)) %>%
  select(Sample, Term, Estimate, `Std. error`) %>%
  nice_table(caption = "Table 9. Two-way fixed effects on a synthetic aimag panel (clustered by aimag)")

Table 9. Two-way fixed effects on a synthetic aimag panel (clustered by aimag)
Sample	Term	Estimate	Std. error
All 22 units	Digital intensity (beta)	0.111***	(0.009)
All 22 units	Digital intensity x herding share (theta)	0.136***	(0.011)
Without Ulaanbaatar	Digital intensity (beta)	0.094***	(0.009)
Without Ulaanbaatar	Digital intensity x herding share (theta)	0.158***	(0.012)

With only 22 clusters, real estimates will need small-sample corrections (for example wild cluster bootstrap), and I will report results with and without the capital (Cameron & Miller, 2015).

3.8 Mapping results to the propositions

Proposition	Finding that supports it	Finding that contradicts it
P1 (deepening)	Positive association of digital uses with formal saving, borrowing and savings-based resilience in Models A and C	Coefficients near zero on all three outcomes
P2 (geography)	Positive \(\theta\) in Model B; larger effects in more herding-intensive aimags in Model C	\(\theta\) zero or negative; larger effects in Ulaanbaatar
P3 (substitution / debt-only)	Digital use rises with little or no change in saving or resilience; or association with borrowing only, especially consumption borrowing	Saving and resilience rise together with borrowing

The three findings are not mutually exclusive. A plausible result is that digital payments are associated with more saving and resilience in rural areas but only with more consumer borrowing in the capital, which would be a partial deepening. That outcome would still be informative: it would show where the stepping-stone logic of Agarwal and Assenova (2024) holds once access is universal.

3.9 Limitations and threats to validity

Ceiling and saturation. Near-universal accounts and near-universal digital merchant payments leave little variation in the latest waves; depth measures and pooled waves are needed, and results are driven by earlier waves.
Selection and reverse causality. People who already save or borrow may use digital payments more. Controls, stability bounds and fixed effects reduce but do not remove this problem, so results are described as associations.
A common cause: government payments. Wages, pensions and child money arrive in accounts and may drive both account use and digital use, which is why they enter as controls and why the Child Money Programme is treated as context rather than an instrument.
Small samples. About 1,000 respondents per wave makes rural subsamples small; I will report confidence intervals and avoid over-reading point estimates.
Repeated cross-sections. The Findex does not follow individuals, so the sequence from payment use to later saving or credit cannot be observed directly.
Geographic resolution. The Findex may not identify the aimag, so provincial variation depends on Bank of Mongolia and NSO publications and may be unavailable.
Nomadic households. Herding share is a proxy for mobility, and highly mobile households may be under-covered in surveys.
Definitions across sources. Central-bank statements on digital payment use and Findex indicators measure different things and should not be mixed in one series.

3.10 Next steps and how to run the real data

Request the Mongolia Findex microdata for 2011, 2014, 2017, 2021 and 2024 from the World Bank Microdata Library (World Bank, 2025a) and confirm the variable names and the coding of urbanicity.
Inventory Bank of Mongolia and NSO series for provincial coverage; confirm the milestone dates in Table 3 from Mongolian-language sources.
Replace syn with the real analysis data frame using the adapter below, and re-run Sections 3.7.1 to 3.7.4 unchanged.

The adapter is a template that has not been run on the real file; every line marked CHECK must be verified against the data dictionary.

# --- Real-data adapter (template; NOT evaluated) ---------------------------------
# raw <- readr::read_csv("MNG_2024_FINDEX_v02_M.csv")      # or haven::read_dta("....dta")

first_present <- function(df, candidates) {
  hit <- intersect(candidates, names(df))
  if (!length(hit)) stop("None of these columns found: ", paste(candidates, collapse = ", "),
                         ". Check the data dictionary.")
  hit[1]
}
yes <- function(x) as.integer(x == 1)    # CHECK: raw Findex yes/no items are usually 1 = yes, 2 = no

build_analysis_df <- function(raw) {
  raw %>% transmute(
    wave   = year,                                           # CHECK
    rural  = as.integer(urbanicity == 1),                    # CHECK coding with table(raw$urbanicity)
    female = as.integer(female == 1),                        # CHECK
    age, educ, inc_q, emp = as.integer(emp_in == 1),         # CHECK
    account = as.integer(account == 1),
    gov_transfer = pmax(yes(receive_transfers), yes(receive_pensions), na.rm = TRUE),
    formal_saving    = yes(fin17a),
    formal_borrowing = yes(fin22a),
    emergency_savings = as.integer(fin24 == 1),
    digital_payment = .data[[first_present(raw, c("anydigpayment"))]],       # CHECK name
    digital_uses = yes(fin3)   # + merchant-payment and bill-payment items: CHECK names, then add here
  ) %>% filter(!is.na(wave))
}
# syn <- build_analysis_df(raw)   # then re-run the models above

References

Agarwal, A., & Assenova, V. A. (2024). Mobile money as a stepping stone to financial inclusion: How digital multisided platforms fill institutional voids. Organization Science, 35(3), 769–787. https://doi.org/10.1287/orsc.2022.16562

Ahmad, A. H., Green, C., & Jiang, F. (2020). Mobile money, financial inclusion and development: A review with reference to African experience. Journal of Economic Surveys, 34(4), 753–792.

Bachas, P., Gertler, P., Higgins, S., & Seira, E. (2021). How debit cards enable the poor to save more. Journal of Finance, 76(4), 1913–1957. https://doi.org/10.1111/jofi.13021

Bank of Mongolia. (2024a). 100 years of the modern banking system in Mongolia [Article by the Governor, L. Byadran]. https://www.mongolbank.mn/en/r/7224

Bank of Mongolia. (2024b). Ten remarkable milestones celebrating 100 years of modern banking in Mongolia. https://www.mongolbank.mn/en/r/8563

Bank of Mongolia. (2025). Monetary policy statement 2025-02. https://www.mongolbank.mn/en/r/8942

Bank of Mongolia. (2026). Monetary policy committee statement 2026-01. https://www.mongolbank.mn/en/r/11422

Beck, T., Pamuk, H., Ramrattan, R., & Uras, B. R. (2018). Payment instruments, finance and development. Journal of Development Economics, 133, 162–186. https://doi.org/10.1016/j.jdeveco.2018.01.005

Bessir, M., Ganbat, E., Klapper, L., & Norris, A. (2026, March 19). Mongolia’s child money program: Early account ownership and financial inclusion at scale. World Bank Blogs, All About Finance. https://blogs.worldbank.org/en/allaboutfinance/mongolia-s-child-money-program--early-account-ownership-and-fina

Borjigin, W., He, S., & Xu, H. (2025). The role of digital financial inclusion in promoting common prosperity: Evidence from Inner Mongolia. International Review of Financial Analysis, 100. https://doi.org/10.1016/j.irfa.2025.103996

Cameron, A. C., & Miller, D. L. (2015). A practitioner’s guide to cluster-robust inference. Journal of Human Resources, 50(2), 317–372.

Chen, P.-F., & Tran, M.-L. (2025). Assessing the impact of payment technology, financial inclusion, and institutions on entrepreneurship: Evidence from middle-income Asian economies. Applied Economics.

Delgerjargal, D. (2024). Factors influencing the intention to use digital wallet. Business and Innovation, 63–77. National University of Mongolia, Business School.

Demirgüç-Kunt, A., Klapper, L., Singer, D., & Ansar, S. (2022). The Global Findex Database 2021: Financial inclusion, digital payments, and resilience in the age of COVID-19. World Bank. https://doi.org/10.1596/978-1-4648-1897-4

Demirgüç-Kunt, A., Klapper, L., Singer, D., Ansar, S., & Hess, J. (2018). The Global Findex Database 2017: Measuring financial inclusion and the fintech revolution. World Bank. (Mongolia series accessed through the World Bank Global Findex database, https://globalfindex.worldbank.org)

Dupas, P., Karlan, D., Robinson, J., & Ubfal, D. (2018). Banking the unbanked? Evidence from three countries. American Economic Journal: Applied Economics, 10(2), 257–297. https://doi.org/10.1257/app.20160597

Finextra. (2024). Trade & Development Bank of Mongolia introduces Apple Pay. https://www.finextra.com/pressarticle/103545/trade--development-bank-of-mongolia-introduces-apple-pay

Firpo, J. (2009, May 12). Mongolia’s landscape equals mobile money opportunity. World Bank Private Sector Development Blog. https://blogs.worldbank.org/en/psd/mongolia-s-landscape-equals-mobile-money-opportunity

FinTech Futures. (2021, July 8). Bank of Mongolia goes live with Compass Plus payments tech. https://www.fintechfutures.com/paytech/bank-of-mongolia-goes-live-with-compass-plus-payments-tech

gogo.mn. (n.d.). SocialPay digital wallet and international QR payments [News article]. https://mongolia.gogo.mn/r/ym6km

International Monetary Fund. (2025). Mongolia: 2025 Article IV consultation (IMF Country Report No. 25/265). https://www.imf.org/en/news/articles/2025/09/15/pr-25298-mongolia-imf-executive-board-concludes-2025-article-iv-consultation

International Monetary Fund. (2026). Mongolia: Staff concluding statement of the 2026 Article IV mission (June 26, 2026). https://www.imf.org/en/news/articles/2026/06/26/mcs062626-mongolia-staff-concluding-statement-of-the-2026-article-iv-mission

International Monetary Fund. (2026). Outstanding loans from commercial banks to households for Mongolia [Financial Access Survey series MNGFCSODCHGGDPPT]. Retrieved from FRED, Federal Reserve Bank of St. Louis. https://fred.stlouisfed.org/series/MNGFCSODCHGGDPPT

Jack, W., & Suri, T. (2014). Risk sharing and transactions costs: Evidence from Kenya’s mobile money revolution. American Economic Review, 104(1), 183–223. https://doi.org/10.1257/aer.104.1.183

Khanna, T., & Palepu, K. (1997). Why focused strategies may be wrong for emerging markets. Harvard Business Review, 75(4), 41–51.

Klapper, L., Singer, D., Starita, L., & Norris, A. (2025). The Global Findex Database 2025: Connectivity and financial inclusion in the digital economy. World Bank. https://doi.org/10.1596/978-1-4648-2204-9

Li, H., Wei, X., & Chen, W. (2025). Digital financial inclusion’s impact on farmers’ income and spatial spillover effects: Evidence from Inner Mongolia, China. Heliyon, 11(3), e42155. https://doi.org/10.1016/j.heliyon.2025.e42155

Oster, E. (2019). Unobservable selection and coefficient stability: Theory and evidence. Journal of Business & Economic Statistics, 37(2), 187–204.

Oxford Business Group. (2013). A crowded retail market: Competition gets more intense as banks expand access and services (The Report: Mongolia 2013). https://oxfordbusinessgroup.com/reports/mongolia/2013-report/economy/a-crowded-retail-market-competition-gets-more-intense-as-banks-expand-access-and-services

Rochet, J.-C., & Tirole, J. (2003). Platform competition in two-sided markets. Journal of the European Economic Association, 1(4), 990–1029. https://doi.org/10.1162/154247603322493212

Suri, T., Bharadwaj, P., & Jack, W. (2021). Fintech and household resilience to shocks: Evidence from digital loans in Kenya. Journal of Development Economics, 153, 102697. https://doi.org/10.1016/j.jdeveco.2021.102697

Suri, T., & Jack, W. (2016). The long-run poverty and gender impacts of mobile money. Science, 354(6317), 1288–1292. https://doi.org/10.1126/science.aah5309

World Bank. (2025a). Global Findex Database 2025, Mongolia 2024 microdata (Reference ID MNG_2024_FINDEX_v02_M) [Data set]. World Bank Microdata Library. https://doi.org/10.48529/72yg-qg77 (catalog: https://microdata.worldbank.org/catalog/7947)

World Bank. (2025b). Saving at a financial institution in the past year, Mongolia [Global Financial Development series DDAI06MNA156NWDB]. Retrieved from FRED, Federal Reserve Bank of St. Louis. https://fred.stlouisfed.org/series/DDAI06MNA156NWDB

Stepping Stones Across the Steppe: Digital Payments and Financial Inclusion in Mongolia

Midterm Project: Investment Analytics and Technology Application

Egshiglen Baatar /114035116/

9th of November 2026