Electronic credit transfers, ATM deposits and card payments at POS lifted online transfer volumes in Kosovo between 2018 and 2024, whereas paper transfers, EFTPOS and cash withdrawals (ATM and POS) depressed them; the analysis uses official monthly data and HAC‑GMM estimation but lacks a clearly exogenous source for causal identification.

Main Finding

Digital, fully electronic channels—especially electronic credit transfers (CTEF), ATM deposits (ATMD), and card payments at POS (CPPOS)—significantly increase the monetary volume of online transfers (used as a proxy for fintech performance) in Kosovo (2018–2024). By contrast, paper credit transfers (CTPF), EFTPOS transactions, ATM cash withdrawals (ATMW), and cash withdrawals at POS (CWPOS) are associated with statistically significant negative effects on online transfer volumes. Credit transfers via ATMs (CTATMs) show no statistically significant effect.

Key Points

• Dependent variable: Online Transfers (OT), measured as transaction value (millions of euros).
• Positive, significant drivers of OT: CTEF (electronic credit transfers), ATMD (ATM deposits), CPPOS (card payments at POS).
• Negative, significant drivers of OT: CTPF (paper credit transfers), EFTPOS (card terminals at merchant POS requiring physical presence), ATMW (ATM cash withdrawals), CWPOS (cash withdrawals at POS).
• Non-significant: CTATMs (credit transfers via ATMs).
• Framing theories: Technology Acceptance Model (TAM) and Diffusion of Innovations (DOI) used to interpret why fully digital, low-complexity channels raise online activity while cash-dependent or physically anchored channels hinder it.
• Policy recommendations by the authors: broaden card acceptance and electronic transfers, connect ATM deposit functionality with mobile banking, pricing nudges to discourage cash use, and financial literacy programs.

Data & Methods

• Data: Official monthly transaction-value series from the Central Bank of Kosovo covering January 2018–December 2024. Transaction categories: CTPF, CTEF, EFTPOS, ATMW, ATMD, CTATMs, CWPOS, CPPOS.
• Outcome: Volume (value) of online transfers in millions of euros (not transaction counts).
• Preprocessing: Variables identified as I(1) were differenced; structural breaks modeled for COVID-19/post-pandemic periods and for 2022 e-payment regulatory reforms.
• Estimation techniques:
  • Ordinary Least Squares (OLS) for baseline comparisons.
  • HAC-GMM (Heteroskedasticity and Autocorrelation Consistent Generalized Method of Moments) used to address endogeneity, autocorrelation, heteroskedasticity, and dynamic relationships in the monthly series.
• Contributions in methods: transaction-level, multi-channel specification and the application of HAC-GMM to produce robust estimates in presence of serial correlation, heteroskedasticity, and potential endogeneity.

Implications for AI Economics

• Measurement & modeling:
  • The paper demonstrates benefits of rich, transaction-level time series for estimating digital-payment impacts—data types well suited for combining causal econometrics (HAC-GMM) with machine‑learning (ML) time-series forecasting or panel methods in AI‑economics research.
  • Structural-break treatment (COVID, 2022 reforms) is crucial; AI/ML models used for policy evaluation should include regime-detection or change-point methods to avoid biased inferences.
• Policy targeting & personalization:
  • ML/AI can help micro-target interventions (e.g., financial-literacy nudges, pricing incentives) to user segments most likely to switch from cash to digital channels, amplifying the positive channels identified (CTEF, ATMD, CPPOS).
  • Intelligent routing and UX personalization (driven by AI) can reduce perceived complexity—addressing TAM/DOI determinants—to accelerate diffusion of fully digital instruments.
• Operational and regulatory applications:
  • AI can enhance fraud detection and cybersecurity—addressing barriers to digital adoption flagged in the paper (trust, privacy concerns).
  • Regulators can apply anomaly detection and causal discovery tools to monitor real-time effects of reforms (such as the 2022 e‑payment changes) and better evaluate intervention effectiveness.
• Cautions & equity considerations:
  • Structural and behavioral barriers (cash dependency, infrastructure gaps) limit digital adoption—purely algorithmic solutions risk widening inclusion gaps if not paired with offline investments (connectivity, literacy).
  • Data privacy and fairness concerns arise when using transaction-level data for AI personalization or enforcement; transparency and safeguards are necessary.
• Research opportunities:
  • Combine HAC-GMM causal estimates with ML-based heterogeneity analysis to identify which population segments yield the largest welfare or GDP-per-capita gains from digital-payment adoption.
  • Use the dataset (or similar central-bank transaction series) to train hybrid models: causal econometrics to estimate policy effects and ML for short-term forecasting and targeted interventions.

Overall, the paper supplies empirical grounding and transaction-level targets that AI-enabled policy tools and economic models can build on—while reminding researchers and policymakers to account for infrastructure, behavioral, and equity constraints when designing AI-driven fintech deployments.