Academic Foundations

Methodology

Every weight, formula, and analytical choice in TGFI is grounded in peer-reviewed literature. 30 papers across 6 dimensions, following the OECD Handbook on Constructing Composite Indicators (2008) framework.

3-Layer Analytical Framework

Data Collection

Gather, clean, classify raw inputs from text and structured sources

Academic basis: DIKW Pyramid (Rowley 2007), Gentzkow/Kelly/Taddy (JEL 2019) Text as Data, CAMEO/GDELT event coding

Article ingestion from Tier 1-3 sourcesLLM-as-Classifier with forced citation2-round fact-check verificationHard data API ingestion (OECD SDMX, GDELT)

Unearth the Numbers

Quantitative analysis, scoring, normalization, and composite blending

Academic basis: OECD Handbook on Composite Indicators (2008), Chinn-Ito KAOPEN PCA methodology, BACE (Sala-i-Martin et al. AER 2004)

Text score aggregation (Tier x Confidence x Recency weights)Hard data normalization (min-max to [-100, +100])Composite blending (bucket-specific text/hard weights)Cross-method convergence measurement

Synthesis

Macro interpretation, cross-bucket patterns, narrative generation

Academic basis: Heuer & Pherson (2010) Structured Analytic Techniques, CIA Tradecraft Primer (2009), GEOII multi-dimensional synthesis

Cross-bucket convergence detectionCross-pair triangulation analysisEvent-driven narrative linkingAnomaly flagging (convergence < 0.3)

Per-Dimension Weight Justification

Trade

30/70

Trade is the most data-rich bucket. OECD bilateral trade statistics are available monthly. Gravity model literature demonstrates flows are well-explained by structural economic variables.

Text Weight: 30%

Hard Data: 70%

OECD Monthly International Merchandise Trade (SDMX API)

Key References

Anderson & van Wincoop (2003). “Gravity with Gravitas.” AER.

Aiyar, Malacrino & Presbitero (2024). “Investing in Friends.” European J. Political Economy.

Investment

40/60

Investment data is quarterly (not monthly), creating more lag. CFIUS/EU screening decisions appear in text before statistics.

Text Weight: 40%

Hard Data: 60%

OECD FDI Statistics by partner, OECD FDI Regulatory Restrictiveness Index

Key References

Kalinova, Palerm & Thomsen (2010). “OECD FDI Restrictiveness Index.” OECD Working Paper.

Mistura & Roulet (2019). “The Determinants of FDI.” OECD Working Paper.

Technology

60/40

Export controls are policy signals first. Patent data lags 18+ months. Technology cooperation/conflict is narrative-driven.

Text Weight: 60%

Hard Data: 40%

WIPO patent filings, OECD MSTI, US BIS Entity List

Key References

Jinji & Ozawa (2024). “Economic Consequences of US-China Decoupling.” CEPR/RIETI.

Gentzkow, Kelly & Taddy (2019). “Text as Data.” J. Economic Literature.

Finance

25/75

Finance is the most data-rich bucket after Trade. Exchange rates, reserves, and capital flows are available at daily-to-monthly frequency.

Text Weight: 25%

Hard Data: 75%

IMF COFER, BIS Triennial Survey, SWIFT RMB Tracker

Key References

Chinn & Ito (2006). “KAOPEN Financial Openness Index.” J. Development Economics.

Cipriani, Goldberg & La Spada (2023). “Financial Sanctions, SWIFT, and the International Payment System.” J. Economic Perspectives.

Leverage

80/20

No bilateral leverage index exists. Weaponization is inherently about threats and actions, which are text events. Hard data measures vulnerability, not weaponization itself.

Text Weight: 80%

Hard Data: 20%

USGS Mineral Summaries, IEA Energy Security, Eurostat CRM

Key References

Farrell & Newman (2019). “Weaponized Interdependence.” International Security.

Clayton, Maggiori & Schoar (2024). “A Theory of Economic Coercion and Fragmentation.” BIS Working Paper.

Policy

50/50

Policy is dual-natured: announcements are text, but GDELT event counts and UN voting alignment provide structured hard data.

Text Weight: 50%

Hard Data: 50%

GDELT 2.0 Events, UN Ideal Point Distance, Treaty databases

Key References

Baker, Bloom & Davis (2016). “Measuring Economic Policy Uncertainty.” QJE.

Caldara & Iacoviello (2022). “Measuring Geopolitical Risk.” AER.

Cooperation-Conflict Spectrum

All scores range from -100 (strong conflict) to +100 (strong cooperation). This bilateral spectrum derives from the CAMEO/Goldstein tradition of event coding (Goldstein, JCR 1992) and extends it to article-level analysis via LLM classification.

-100

+100

Strong ConflictNeutralStrong Cooperation