Institutional Capture: A Quantitative Framework for Detection

Introduction

Institutional capture — the progressive displacement of an organization's stated mission by the private interests of those who control it — remains a persistent feature of large Latin American and global enterprises. Its costs are well documented in case studies, but its detection has largely depended on retrospective narrative reconstruction rather than on structured, quantitative signals.

This article proposes an operational framework for approaching detection quantitatively. The framework is composed of four signals — G1, S1, F1, and P3 — and a composite perception index with an explicit aggregation formula. The goal is modest: to give regulators, boards, investors, and forensic analysts a common vocabulary for discussing capture, together with a reproducible starting point for scoring specific cases.

This is the first entry in a three-part series on governance risk analytics. A companion piece — Turning Risk Perception Into Investment Opportunity (published 13 May 2025) — describes instruments of action once perceived risk materializes; a third article, Running Is Not Testing — and Testing Is Not Defending, distinguishes preventive stress testing from forensic analysis under adversarial scrutiny. The present piece should be read as a work in progress: conceptual framework with known methodological gaps that the series is built to close. Peer feedback is welcomed.

---

Framework: G1, S1, F1, P3 — operationalized

Four signals are proposed, each with a declared measurement approach, data source, and range.

G1 — Governance signal. Measured as the ratio of loyalty-based to merit-based executive appointments over a defined rolling window (typically five years). Proxies: board turnover rate, share of directors with declared political affiliation, share of appointments without a competitive selection record. Sources: public CVs, board minutes, sector press. Range: 0 (purely meritocratic) to 1 (purely loyalty-based).

S1 — Strategic signal. Measured as the divergence between declared strategic objectives (annual reports, investor presentations) and observed operational allocation (CAPEX, procurement, hiring, geographic expansion). Proxies: non-core CAPEX share, expansion into high-risk jurisdictions without documented return thesis, related-party procurement share. Range: 0 (full alignment) to 1 (maximum divergence).

F1 — Financial signal. Measured as the discrepancy between reported financial performance and sector benchmarks under comparable conditions. Proxies: residual operating margin after sector controls, working-capital anomalies, concentration of related-party transactions. Range expressed in standard deviations from sector-peer median.

P3 — Perception composite. An aggregate of four sub-indicators drawn from the public information environment around the entity. The explicit formulation:

P3 = (w1·Sentiment + w2·Credibility + w3·Context + w4·Volume) / Σwᵢ

Where:

• Sentiment ∈ [-1, 1] — polarity score of reports, whistleblower testimony, and strike-related coverage. Scoring method must be declared (options: VADER, LLM-scored with specified prompt, or human-coded with inter-rater agreement reported). For aggregation in P3, the magnitude |Sentiment| is used: polarity directs interpretation (negative = adverse coverage; positive = mimetic boosterism), but magnitude is what contributes to the alert signal — strongly negative and strongly positive sentiment both indicate elevated public attention worth escalating.
• Credibility ∈ [0, 1] — share of reports containing specificity markers (named actors, dated events, quantified amounts) over total reports in the window.
• Context ∈ [0, 1] — share of reports with independently verifiable cross-references.
• Volume — normalized log(n_reports), scaled across a defined temporal window (typically 12–36 months).

Initial weights, proposed as hypotheses for calibration:

Weight	Component	Initial value	Rationale
w1	Sentiment	0.20	Directionally informative but easily inflated by mimetic coverage
w2	Credibility	0.35	Specificity is the strongest single filter against noise
w3	Context	0.30	Independent corroboration anchors claims
w4	Volume	0.15	Necessary but weakly specific on its own

These weights are initial hypotheses, not empirically calibrated. Calibration against a labeled corpus of confirmed and rejected capture cases is the next methodological step (see Future Work).

---

Thresholds and regime classification

Three operational regimes are proposed, with deliberately heuristic thresholds:

• P3 ≥ 0.75 — high alert. Capture is consistent with the observed perception signal.
• P3 ∈ [0.45, 0.75) — watch. Ambiguous regime; signal alone is not informative.
• P3 < 0.45 — low alert. Background level of institutional noise.

These thresholds are heuristic operational points derived from illustrative cases. They are not statistically validated: establishing robust cutoffs would require a labeled dataset of n ≥ 50 cases, which is the scope of future work. Users of the framework should treat them as starting anchors, not decision rules.

---

Case 1 — PDVSA (Venezuela)

Two distinct periods anchor the PDVSA case: the capture period (2002–2010) and the collapse period (2014–2020). Production fell from approximately 3.5 million barrels per day in the late 1990s to under 400,000 barrels per day by 2020 (CFR 2022). Reports of diverted resources in the 2004–2014 range (Giordani 2014, Transparencia Venezuela 2023) combine with the 2002 mass dismissal of technical staff (approximately 19,000 employees) to produce a P3 calculation, over the 2005–2010 window, as follows:

Component	Value	Note
Sentiment	−0.80 (|·| = 0.80)	Highly negative sentiment in specialized press; magnitude used in aggregation
Credibility	0.85	Named actors, dated events, quantified amounts present
Context	0.70	Cross-verified across CFR, Transparencia, academic work
Volume (norm.)	0.80	Sustained high reporting volume
P3	0.79	High alert regime

Explicit arithmetic for reproducibility: P3 = (0.20·0.80 + 0.35·0.85 + 0.30·0.70 + 0.15·0.80) / (0.20 + 0.35 + 0.30 + 0.15) = (0.160 + 0.2975 + 0.210 + 0.120) / 1.00 = 0.7875 ≈ 0.79. Above the 0.75 high-alert threshold. Any third-party reproducing the framework on the same component values must reach P3 ∈ [0.78, 0.80] within rounding tolerance.

The G1 signal for the same window was elevated (loyalty-based appointment share above 0.7 on declared-affiliation proxies), and S1 diverged materially as non-core CAPEX grew. The subsequent F1 collapse is consistent with — though not proven to be caused by — the earlier governance and strategic signals.

Case 2 — Petrobras (Brazil)

The Petrobras case offers a contrast: elevated signals during the 2010–2014 window, followed by material signal reduction after the 2014–2020 Operação Lava Jato judicial and institutional response. P3 during the active capture phase lands around 0.60–0.65 on comparable scoring (high-volume credible reporting, partially verifiable via public prosecutors' disclosures). Post-2016, production rose approximately 20% between 2016 and 2018 (Reuters 2018), with concurrent board reforms and compliance overhaul. The Petrobras trajectory is compatible with the rescue pattern, distinguished from sustained capture in the next section.

Cases 3–5 — expanding the sample

Three additional public cases are offered, briefly, to move beyond n = 2. Full P3 computations for these cases will accompany the next methodological entry; summary readings follow.

Pemex (Mexico, 2012–2022). Chronic governance and strategic divergence without a clearly observable rescue phase. Preliminary P3 readings are consistently in the high-alert range; F1 deterioration is sustained rather than cyclical.

YPF (Argentina, 2012–2019). The 2012 nationalization episode combined acute governance shocks with strategic realignment. P3 peaked during 2013–2015, then reverted partially as operational metrics stabilized. The case illustrates the importance of distinguishing transient from chronic capture.

Sonangol (Angola). Post-colonial state-owned enterprise with long-dated governance concerns and a more recent reform cycle (2017 onward). Useful as a non-Latin-American reference that nevertheless matches the LATAM pattern in its primary signals.

These cases are included to illustrate framework transferability. They are not a representative sample: selection bias toward visible cases is acknowledged.

---

Capture vs. rescue — differentiation

Not every high-alert regime persists. The framework distinguishes sustained capture from rescue trajectories through three criteria over a 3–5 year post-alert window:

• F1 normalization toward sector peer median.
• S1 realignment between declared and observed strategic allocation.
• G1 substitution pattern — replacement of appointment patterns rather than cosmetic reshuffling.

When all three are observed, the entity's trajectory is consistent with a rescue pattern (the Petrobras 2016–2018 period is an approximate example). When none or only one is observed, the trajectory is consistent with continued capture (Pemex across the reference window).

---

Limitations

This framework is a working methodology, not a validated inference tool. The following limitations are acknowledged:

1. Sample size. The five cases presented are insufficient for statistical inference. Robust calibration requires n ≥ 50 labeled cases.
2. Method dependence. Each P3 component admits several measurement approaches. Different methods produce different numerical results on the same corpus.
3. Weight calibration. The initial weights (0.20 / 0.35 / 0.30 / 0.15) are hypotheses, not empirically derived.
4. Threshold fragility. The 0.75 and 0.45 cutoffs are heuristic operational anchors, not statistically derived decision rules.
5. Confounding. Institutional capture and macroeconomic crisis frequently co-occur. Causal attribution would require instrumental variables not available in the public data used here.
6. Selection bias. Cases are drawn from visible public controversies, not from a representative sample of state-owned or privately held enterprises.
7. Retrospective orientation. The framework currently operates on historical windows. Prospective early-warning variants require further development.

---

Future work

The third article in the series will formalize the P3 calibration protocol — specifically, the labeled corpus construction, inter-rater agreement procedure, and weight derivation via supervised learning. Additional steps on the roadmap: expansion to n ≥ 50 cases, publication of a reproducible computation pipeline (code and data sources), validation against procurement red-flag and labor-strike datasets, and development of a prospective variant with rolling-window updates.

---

Conclusion

The G1 / S1 / F1 / P3 framework offered here is a common vocabulary and a heuristic operational tool. It is not, in its current form, a statistically robust inference method, and this article does not claim otherwise. Its value lies in making the conversation about capture tractable — letting regulators, boards, and analysts compare cases with a shared structure rather than anecdote. The methodological gaps acknowledged above are the work program of the series.

Peer discussion, case contributions, and methodological critique are welcomed.

---

Glossary

• Institutional capture — the progressive substitution of an organization's stated objectives by the private interests of those who control it.
• G1 (governance signal) — ratio of loyalty-based to merit-based executive appointments.
• S1 (strategic signal) — divergence between declared and observed strategic allocation.
• F1 (financial signal) — discrepancy between reported performance and sector benchmarks.
• P3 (perception composite) — weighted aggregate of Sentiment, Credibility, Context, and Volume over a defined reporting window.
• Regime — discretization of P3 into high-alert, watch, and low-alert bands.
• Rescue trajectory — post-alert trajectory showing F1 normalization, S1 realignment, and G1 substitution within 3–5 years.

---

References

• Carpenter, D. (2014). Reputation and Power. Princeton University Press.
• Council on Foreign Relations (2022). Venezuela: The Rise and Fall of a Petrostate.
• Dal Bó, E. (2006). Regulatory capture: A review. Oxford Review of Economic Policy, 22(2).
• EL PAÍS (2020). Coverage of Operação Lava Jato outcomes and Petrobras governance reforms.
• Giordani, J. (2014). Testimony on resource allocation during the 2004–2014 period.
• Reuters (2018). Coverage of Petrobras production recovery, 2016–2018.
• Stigler, G. (1971). The theory of economic regulation. Bell Journal of Economics, 2(1).
• Transparencia Venezuela (2023). Empresas del Estado: Informe Anual.
• Transparency International (2023). Latin America & Caribbean regional corruption reports.