📈 economics

A Multi-LLM Consensus Weighting (MCDM) Approach to Redefining Microfinance Through Blockchain: Enriching Social Capital Outreach in Malaysia and the Maldives

This study employs a novel Multi-LLM Consensus Weighting method to evaluate and rank the social capital conditions for blockchain-enabled microfinance in Malaysia and the Maldives against 53 OIC economies, demonstrating robust comparative insights that inform policy on financial inclusion, transparency, and digital readiness.

Original authors: Zahari Md Ro, Nur Aima Shafie, Zuraidah Md San, Aizatul Waznah Ghazali, Mariyam Niyaf Mohamed

Published 2026-07-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Zahari Md Ro, Nur Aima Shafie, Zuraidah Md San, Aizatul Waznah Ghazali, Mariyam Niyaf Mohamed

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to build a bridge to help people who have been left behind by the big banks. In the world of finance, this is called microfinance: giving small loans to people who need them to start a business or buy food, but who can't get a loan from a regular bank because they don't have a fancy house or a big savings account to use as collateral. For these loans to work, the community needs social capital. Think of social capital as the "glue" of a neighborhood: it's the trust, the honesty, and the way neighbors look out for each other. If everyone trusts each other, they can lend money without needing a lawyer or a bank vault.

Now, imagine a new tool called blockchain. You can think of blockchain as a super-secure, digital notebook that everyone can see but no one can erase or cheat on. It records every single transaction perfectly. The big question researchers are asking is: Can this digital notebook make the "glue" of the neighborhood even stronger? Can it help microfinance reach even more people by making things more transparent and trustworthy? This is the puzzle a team of researchers set out to solve, specifically looking at how well different countries are ready to use this technology to help their poorest citizens.

The Digital Scorecard: How Malaysia and the Maldives Stack Up

The researchers, led by Zahari Md Rodzi and his team, decided to build a "report card" to measure how ready different countries are for this blockchain-powered future. They didn't just look at one thing; they looked at the three ingredients needed to make the magic happen: Transparency (how honest and open the government and institutions are), Financial Inclusion (how many people actually have bank accounts), and Digital Readiness (how many people have internet access).

To grade these countries, they used a clever new method called Multi-LLM Consensus Weighting (MLCW). Here's the fun part: instead of just asking a few human experts to decide which ingredient is most important (which can be slow and biased), they asked a panel of five different Artificial Intelligence "experts" (Large Language Models) to weigh in. They treated these AIs like a jury. If the AI jury agreed on a point, that point got a strong vote. If they disagreed, the system automatically lowered the weight of that point to avoid bias. Then, they mixed this "AI opinion" with hard math based on how much the data varied across countries. It's like having a wise council of robots and a strict math teacher working together to make the final grade.

They tested this system on 53 countries in the Organisation of Islamic Cooperation (OIC), but they focused the spotlight on two specific collaborators: Malaysia and the Maldives.

The Results:
When the dust settled, the report card revealed a clear picture of the landscape:

Malaysia scored an impressive 2nd place out of 53 countries, with a score of 0.850. They are a powerhouse, combining high trust, almost everyone having a bank account, and near-universal internet access. For Malaysia, the challenge isn't getting people into the system; it's about using blockchain to go deeper and help people with "thin" credit histories who are still left out.
The Maldives came in 9th place with a score of 0.698. They are doing well on having bank accounts, but they struggle a bit more with transparency and the scattered nature of their islands makes it hard to monitor things. For the Maldives, blockchain could be a game-changer for creating a transparent, unbreakable record of who owes what, solving their specific problem of trust across the ocean.

The study found that the rankings were incredibly stable. Whether they used the AI panel, just the math, or even a simple "equal weight" method, Malaysia stayed near the top and the Maldives stayed in the middle. The researchers even ran the numbers 2,000 times using a technique called "bootstrap resampling" (basically shuffling the deck of countries over and over to see if the results hold up), and the rankings remained solid.

What This Means:
The paper suggests that blockchain isn't a magic wand that fixes everything instantly. Instead, it's a tool that works best when the foundation is already strong. In countries like Malaysia, where the foundation is solid, blockchain can help deepen relationships and reach the very last mile of the poor. In places like the Maldives, where the foundation has some cracks (specifically in transparency), blockchain offers a way to build a "digital audit trail" that makes trust easier to verify without needing expensive physical oversight.

The researchers are careful to say this is a snapshot of potential conditions, not a guarantee of future success. They didn't test the actual loans; they measured the environment needed for the loans to succeed. But by proving that their new "AI-jury" method produces consistent, reliable results, they've given policymakers a new, auditable way to see where their countries stand and what they need to build next. It's a bit like checking the weather and the soil quality before planting a garden; you can't guarantee the harvest yet, but you know exactly what you need to make it grow.

Technical Summary: A Multi-LLM Consensus Weighting (MCDM) Approach to Redefining Microfinance Through Blockchain

Problem Statement
Microfinance outreach—the capacity to extend financial services to underserved households and microenterprises—is fundamentally dependent on social capital (trust, transparency, and connectivity). While blockchain technology is frequently proposed as a mechanism to enhance these social-capital conditions through tamper-evident ledgers and programmable contracts, there is currently no auditable, comparable method to measure the social-capital readiness for blockchain-enabled microfinance across different nations. Existing multi-criteria decision-making (MCDM) approaches often focus on ranking alternatives rather than addressing the critical methodological challenge of weighting criteria. Furthermore, traditional weighting methods suffer from a dichotomy: subjective methods (e.g., AHP, Delphi) are costly and non-reproducible, while objective methods (e.g., Entropy, CRITIC) are reproducible but "meaning-blind," ignoring the semantic importance of indicators in specific contexts like Islamic finance.

Methodology
The study employs a parsimonious indicator framework and a novel weighting technique, Multi-LLM Consensus Weighting (MLCW), to construct a Social Capital Outreach Index (SCI) for Malaysia and the Maldives, benchmarked against 53 Organisation of Islamic Cooperation (OIC) economies.

Indicator Framework: The study utilizes a minimal set of three benefit-type indicators selected for complete international coverage, reliability, and reproducibility:
- Transparency: Proxied by the Corruption Perceptions Index (Transparency International).
- Financial Inclusion: Proxied by account ownership rates (World Bank Global Findex).
- Digital Readiness: Proxied by internet usage rates (ITU/World Bank).
  Data is normalized using min-max scaling over the OIC panel. Missing values are handled via complete-case analysis rather than imputation.
Multi-LLM Consensus Weighting (MLCW): This is the core methodological innovation. It fuses a semantic branch with a statistical branch:
- Semantic Branch: A panel of $M$ Large Language Models (LLMs) acts as independent assessors to judge the conceptual importance of indicators based on domain knowledge (Islamic finance, development economics). To ensure fairness, models are blinded to country data. The models' weight vectors are aggregated using reliability weights (derived from agreement with the panel median) and a disagreement penalty ( $\lambda$ ) that attenuates weights for indicators where the panel is divided.
- Statistical Branch: Objective entropy weights are calculated based on the statistical dispersion of the data across the OIC panel.
- Fusion: The semantic and objective weights are combined via a convex fusion parameter ( $\alpha \in [0, 1]$ ), allowing the index to range from purely objective ( $\alpha=0$ ) to purely semantic ( $\alpha=1$ ). The study primarily reports results at the objective operating point ( $\alpha=0$ ) for maximum reproducibility.
Validation: The study employs non-parametric bootstrapping (2,000 draws) to quantify uncertainty and conducts robustness checks by comparing MLCW results against Equal, Entropy, CRITIC, and MEREC weighting methods.

Key Results

Ranking: Within the OIC panel of 53 economies, Malaysia ranks 2nd (SCI: 0.850) and the Maldives ranks 9th (SCI: 0.698). The top tier is dominated by Gulf Cooperation Council (GCC) economies and Malaysia, characterized by high institutional integrity and near-universal account ownership.
Weighting Convergence: A significant finding is the convergent validity between the semantic and objective branches. For the two-indicator core (Transparency and Financial Inclusion), the LLM panel's semantic weights (0.453, 0.547) are nearly identical to the objective entropy weights (0.455, 0.545). Consequently, the rankings of Malaysia and the Maldives remain stable (2nd and 9th) across all fusion parameters ( $\alpha$ ) and alternative weighting methods.
Robustness: The rankings are highly robust. Pairwise Spearman correlations between rankings generated by Equal, Entropy, CRITIC, and MEREC methods are $\ge 0.997$ . Adding a third indicator (Digital Readiness) to a smaller subset of 19 economies shifts the Spearman correlation to 0.954, with only minor rank adjustments (Malaysia moves to 4th, Maldives to 11th).
Tiered Structure: The OIC economies exhibit a three-tier gradient: a top tier (integrity + inclusion), a middle tier (variable integrity, moderate inclusion), and a lower tier (conflict-affected, low inclusion).

Significance and Contributions
The paper claims three primary contributions:

Theoretical: It reframes the subjective-objective dichotomy in MCDM weighting as a continuous spectrum governed by a single parameter ( $\alpha$ ). It introduces MLCW as a method that replaces human committees with a reproducible, multi-model panel, thereby retaining domain "meaning" while achieving the reproducibility of objective methods.
Empirical: It provides the first auditable, comparative measure of the social-capital conditions for blockchain-enabled microfinance across OIC economies. It demonstrates that a parsimonious, data-driven index can yield method-invariant conclusions, validating the use of such indices for policy benchmarking.
Practical: The index serves as a defensible tool for policymakers and Islamic finance regulators. It highlights that for mature economies like Malaysia, the policy frontier is "depth" (extending reputational credit via blockchain), whereas for economies like the Maldives, the priority is "integrity" (using auditable ledgers to substitute for costly physical oversight in dispersed geographies).

Limitations and Scope
The authors explicitly state that the study is associational, not causal; it characterizes conditions for potential outreach rather than measuring the realized impact of specific interventions. The index relies on national-level proxies and does not capture institution-level metrics (e.g., borrower counts). The headline results use the objective operating point ( $\alpha=0$ ), leaving the activation of the live semantic branch with logged model diagnostics as a future step. The framework is intentionally parsimonious, excluding dimensions like general trust or participation due to a lack of comparable OIC-wide data.

The Digital Scorecard: How Malaysia and the Maldives Stack Up

More like this