VNU Logo
Vietnam National University
UEB Logo
University of Economics and Business
2026 Poster Prize Student Scientific Research Vietnam National University, Hanoi

From Retrospective Benchmarking to Anticipatory Policy: A Data‑Driven Framework for
Governance Indexing and Forecasting Using the Viet Nam Provincial Governance and
Public Administration Performance Index (PAPI)

1. Introduction
Problem Statement
  • Current subnational governance indices (PAPI, PCI) are predominantly retrospective and rely on equal-weighting defaults.
  • This limits capacity to provide forward-looking signals and early warnings for proactive policymaking.
  • A data-driven, anticipatory framework is urgently needed across Vietnam’s 63 provinces.
Research Objectives
  1. Develop a hybrid governance framework combining Two-Level Hierarchical CRITIC Weighting with multi-window temporal stability validation (RMSD, Spearman’s ρ, Kendall’s W, Monte Carlo sensitivity).
  2. Establish a multi-methodological MCDM Consensus Ranking framework (TOPSIS, VIKOR, PROMETHEE II, COPRAS, EDAS) to transcend single-method idiosyncrasy, validate inter-method concordance, and identify an optimal ranking function.
  3. Generate probabilistic projections for 2025 to operationalize an institutional Early Warning System for subnational governance.
2. Methods & Data
Data Description
  • Balanced panel dataset from PAPI (2011–2024) across all 63 provinces of Viet Nam.
  • 2011–2017: 6 criteria (C01–C06), comprising 21 sub-criteria.
  • 2018–2024: Expanded to 8 criteria, adding C07: Environmental Governance and C08: E-Governance, totalling 29 sub-criteria (SC11–SC83).
Three-Tier Methodological Framework
PAPI Longitudinal Panel 2011–2024
TIER 1 — Two-Level Hierarchical CRITIC Weighting Level 1: Sub-criterion weights within criteria · Level 2: Cross-criteria weights capturing contrast intensity
TIER 2 — Multi-Method MCDM Consensus Ranking TOPSIS · VIKOR · PROMETHEE II · COPRAS · EDAS
TIER 3 — AutoGluon Stacked Ensemble Forecasting (2025) 10 diverse base learners: Statistical · Tabular ML · Deep Learning · Chronos foundation model
3. Weighting & Ranking Results
Sub-criterion weight heatmap 2011-2024

Fig. 1 — Sub-criterion weight heatmap (2011–2024)

Table 1 — Mean CRITIC Criterion Weights
Criterion Mean Wt. CV Informational Role
C01: Local Participation
14.21%
0.149 Lowest weight variance (most consistent)
C02: Transparency
12.46%
0.230 Lowest weight among long-standing criteria
C03: Vertical Accountability
13.22%
0.211 Moderate information value
C04: Control of Corruption
13.64%
0.317 Highest temporal weight volatility
C05: Public Admin. Procedures
14.59%
0.213 Highly stable administrative priority
C06: Public Service Delivery
16.76%
0.238 Primary differentiator across provinces
C07: Env. Governance (since 2018)
16.28%
0.142 Graceful integration of index expansion
C08: E-Governance (since 2018)
13.99%
0.157 Graceful integration of index expansion
Key Findings & Validation
Rank stability over time

Fig. 2 — Rank stability over time

Ranking discriminatory power via score inter-quartile range

Fig. 3 — Ranking discriminatory power (Score IQR)

4. Forecasting Results (2025)
0.6411 Ensemble MASE Well below 1.0 naive baseline
+14.12% vs. Best Single (TFT) TFT baseline MASE = 0.7316
+86.0% vs. Naive Baseline Naive MASE = 1.1924
2025 Provincial Governance Forecast Choropleth Map

Fig. 4 — 2025 Provincial Governance Forecast (PROMETHEE II scores)

Leading Cluster
Quang Ninh1.000
Thai Nguyen0.862
Bac Ninh0.855
Ha Tinh0.819
Lagging Cluster
Phu Yen0.000
Kien Giang0.020
Kon Tum0.037
Dak Nong0.060
Metropolitan
Hanoi0.477
Ho Chi Minh0.302

Moderate scores reflect urbanisation-driven service scaling challenges.

AutoGluon ensemble model weight distribution

Fig. 5 — AutoGluon stacked ensemble model family weight distribution

Steepest governance declines forecast

Fig. 6 — Composite PAPI score decline forecast

MASE comparison across models

Fig. 7 — MASE comparison: ensemble vs. baselines

5. Discussion & Policy Implications
Projected Governance Contractions
  • 29 of 63 provinces are projected to experience governance contractions in 2025.
  • Systemic bottlenecks: Internet Access (SC82) and Civic Knowledge (SC11) signal issues in digital administration and grassroots civic engagement.
  • Top declines: Ca Mau (~50%), Dak Lak (~48%), Quang Ngai (41.12%), Binh Thuan (32.55%).
5-Pillar Policy Response Matrix
Policy Pillar Target Sub-Criteria Regulatory & Operational Response
1Digital Access & E-Governance SC82 Internet Access
SC81 E-Gov Portals
  • Ringfence USO capital for last-mile broadband in lagging provinces.
  • Align KPIs with Nat. Digital Transformation Program (Decision 749/QD-TTg).
  • Deploy commune-level digital navigators (Lao Cai model).
2Civic Engagement & Grassroots Democracy SC11 Civic Knowledge
SC14 Voluntary Contrib.
  • Mandate commune deliberative forums (Law 10/2022/QH15, Decree 59/2023).
  • Establish Village Development Funds with transparent community auditing.
3Accountability & Anti-Corruption SC23 Budget Trans.
SC42 Corruption Control
  • Standardize commune budget reports via MoF templates.
  • Deploy targeted SAV audits in high-risk provinces.
  • Mandate e-procurement at district level.
4Environmental & Resource Governance SC73 Water Quality
SC71 Env. Protection
  • Inter-provincial Mekong water quality compact to resolve jurisdictional gaps.
  • Recapitalize forest payments (Decree 156/2018 & 91/2024) in Central Highlands.
5Tiered Governance Compact (TGC) Composite Scores
  • Top-Quartile Provinces: Enhanced fiscal autonomy and reduced compliance burden.
  • Bottom-Quartile / Declining: Trigger central support mechanisms and capacity programs.
6. Conclusion
Academic Contribution
Bridges static MCDM ranking with dynamic temporal ML forecasting, defining a replicable methodology for Anticipatory Governance in developing economies.
Policy Value
Transitions subnational governance from retrospective benchmarking to proactive one-year-ahead early-warning interventions at scale.
Reproducibility
Code repositories and processed datasets are published as open-source assets for full scientific reproducibility and peer replication.