Macro Risk Monitor · 2/4
The Risk Measurement Model
This page publishes all the calculation rules. The goal is that readers can reproduce the same results from the same inputs. The current model version is v2.4.
Contents
Structure and versions
v1.0, the first version, used a simple formula: Score=(0.7M+0.3L+0.25D)×20. v2.0 made the following eight changes.
| Change | Reason |
|---|---|
| Separated direction D from the score | The index fell whenever deterioration merely stopped |
| Limited risk ② to public finances | It double-counted with FCO, which measures financial stress |
| Added both FCO and FFI | Stress alone misses a state like 2007 |
| Made SURI a 20-country measure | An average flattens the distribution |
| Extended back-testing from 2008 alone to five episodes | One point in time cannot check a scale |
| Added anchors, mechanical calculation and a prediction log | Prevents scoring from drifting |
| Moved data to a long format linked by formulas | Prevents manual copying errors |
| GRCS as 0.6 mean + 0.4 top two | Avoids the dilution of a simple mean and the ceiling effect of a maximum |
| Version | When | Content |
|---|---|---|
| v2.1 | September 2026 | Added a sixth FCO term (intervention usage) and six indicators. August final report |
| v2.2 | September 2026 | Full test of 356 months. Changed the model's positioning |
| v2.3 | September 2026 | Recomputed FCO for the episodes |
| v2.4 | October 2026 | September final report. Hypothesis framework. Withdrew the FFI×GRCS4 warning proposal. Rule against L ratcheting |
Index definitions
Level = (0.7 × M + 0.3 × L) × 20
GRCS5 = 0.6 × (mean Level of five risks) + 0.4 × (mean Level of top two risks)
GRCS4 = same formula excluding ⑤ AI and labor
SCI = (mean strength of five chains) × 20
SURI = 0.35 Household Stress + 0.25 Protest Mobilization + 0.20 Government Fragility + 0.20 Conflict Transmission
RCEI = 0.30 Combat Intensity + 0.25 Escalation Speed + 0.15 State Fragility + 0.15 Food, Energy, Water + 0.15 Refugees and Cross-border
FFI = 0.20 Leverage and Credit Quality + 0.15 Maturity Wall + 0.20 Valuation and Spread Compression + 0.20 Opaque Assets + 0.10 Bank Lending and Balance Sheets + 0.15 External Vulnerability
FCO = 0.25 Credit + 0.20 Volatility + 0.15 Funding + 0.15 Banks + 0.10 Liquidity + 0.15 Intervention Usage
| Index | Full name | Use |
|---|---|---|
| GRCS5 | Global Rolling Crisis Score | The main measure of current compound macro pressure |
| GRCS4 | Same, excluding ⑤ | Only for comparison with the past, when no AI data exists |
| FFI | Financial Fragility Index | Financial weaknesses that build up in calm times |
| FCO | Financial Contagion Overlay | Stress occurring now in the private financial system |
| SCI | Systemic Coupling Index | Strength of the five chains |
| SURI | Social Unrest Risk Index | Social unrest in 20 countries. Aggregated as the count ≥60, population-weighted, GDP-weighted and maximum |
| RCEI | Regional Conflict Escalation Index | Escalation of regional conflict |
M and L run from 1 to 5, and direction D from −2 to +2. D is not in the score and is shown separately.
Scoring anchors
M is scored against the table below. L asks whether a path that pushes another risk's M up one step actually exists. As a guide it is the maximum of the related chains. Discretion is limited to ±1, with the reason noted.
| M | ① Climate, food, water | ② Fiscal (public sector) | ③ Energy, shipping | ④ Politics, geopolitics | ⑤ AI, labor |
|---|---|---|---|---|---|
| 1 | Temperature anomaly above +0.3℃, no price effect | Interest/revenue around the past average | Prices and traffic in normal range | Ordinary political conflict | Limited adoption, no effect in employment data |
| 2 | FAO nominal under +3% y/y, local output falls | Long yields above the 70th percentile of the past 10 years | Prices in top 25% of the past 5 years, or traffic down 10%+ | Concern over policy discontinuity from elections or change of government | Hiring restraint in some sectors |
| 3 | FAO real above the 60th percentile of the past 20 years, yields cut 5%+ in two major producers | Interest/revenue above 10% in major economies, wider auction tails | Traffic at a major chokepoint down 30%+, or oil above the 80th percentile of the past 10 years | Sanctions and military tension among major powers, battle events +10% on 3 months | Entry-level postings down 5%+ y/y, AI-attributed layoffs above 20% of total |
| 4 | The above plus export limits in 3+ countries, or stocks-to-use 5pt below the 20-year mean | Spending up and revenue down together in several countries, talk of downgrades | Traffic down 70%+, or supply down 5 mb/d+ | Regional war under way, battle events +25% on 3 months, a major power directly involved | Broad fall in white-collar postings, revenue-employment decoupling in double-digit points |
| 5 | Simultaneous poor harvests in several major grains, export limits in 10+ countries, real index near record | Failed auctions routine in major economies, chain of sovereign downgrades | Several chokepoints closed at once, supply loss above 10 mb/d | Large armed conflicts in several regions at once, direct clash among major powers | White-collar unemployment clearly worse in macro data |
FCO conversions
Credit, volatility and intervention usage are mechanical. Funding, banks and liquidity are scored by expert judgment (40% in total). The conversion is a stress conversion, not a percentile. Values near the median score about 10, and the score rises fast in the upper tail. The breakpoints approximate the 1997-2025 distribution.
HY OAS (%) to score: 2→0, 2.5→2, 3→6, 3.5→12, 4.5→22, 5.5→35, 7→50, 9→68, 12→85, 18→95, 24→100
VIX to score: 10→0, 14→5, 17→10, 20→20, 25→35, 30→50, 40→70, 55→85, 80→100
Intervention stress = MIN(100, MAX(0, 100 × LN(MAX(X, 10) / 10) / LN(45))) X = monthly peak of Fed lending ($ billion)
Values between points are interpolated linearly. Fed lending scores 0 at $10 billion or less, and 100 at $450 billion.
Band thresholds
| Index | Low | Watch | Elevated | High | Severe | Systemic |
|---|---|---|---|---|---|---|
| GRCS | <40 | 40-54 | 55-69 | 70-79 | 80-89 | 90+ |
| FFI | <35 | 35-49 | 50-64 | 65-74 | 75-84 | 85+ |
| FCO | <15 | 15-29 | 30-49 | 50-69 | 70-84 | 85+ |
| SURI, RCEI | <20 | 20-39 | 40-59 | 60-74 | 75-89 | 90+ |
Regime rule
The regime is decided from the top, and the first matching row applies.
| Order | Condition | Regime |
|---|---|---|
| 1 | FCO ≥ 70 and GRCS5 ≥ 80 | Systemic Compound Crisis |
| 2 | FCO ≥ 70 | Financial-Led Crisis |
| 3 | FCO ≥ 30 and GRCS5 ≥ 75 | Macro-Financial Contagion |
| 4 | FCO ≥ 30 | Financial Stress |
| 5 | GRCS5 ≥ 75 and FFI ≥ 65 | Compound Macro Stress / Latent Fragility |
| 6 | GRCS5 ≥ 75 | Compound Macro Stress |
| 7 | GRCS5 ≥ 55 | Elevated Macro Stress |
| 8 | Otherwise | Normal |
Model Validation shows the rule applied to past episodes. We issue a special report in any month the regime changes.
proxy-GRCS
To check our judgment scores, we show alongside them an objective version computed mechanically from market data and statistics alone.
| Axis | Source | Treatment |
|---|---|---|
| Food | IMF PFOODINDEXM | Deflated by CPI |
| Fiscal | ACM 10-year term premium (THREEFYTP10) | As is |
| Energy | IMF POILBREUSDM | Deflated by CPI |
| Geopolitics | GPR index | As is |
Each axis is converted to an expanding-window percentile (60-month burn-in) and aggregated with the GRCS formula. CPI is CPIAUCSL, with October 2025 linearly interpolated because of the government shutdown. If the gap with the judgment-based GRCS4 exceeds ±15, we review the scoring criteria.
Prediction log and Brier score
Each report issues predictions with a probability and a deadline that anyone can judge. The Brier score is the mean of (probability − outcome)². Zero is perfect, lower is better, and answering 50% every time gives 0.25. A guide using base rates is 0.20.
If we do not stay below 0.25 over time, we will review the model's structure and weights. Threshold predictions are settled when reached, and unmet predictions are not counted as 0 until their deadline. Hits therefore come first in the early period. A hit given a low probability still worsens the score, which is the correct behavior.
Data practices
| Status | Meaning |
|---|---|
| Actual | Official statistics or market data used as is |
| Approximate | Rough figures from reports and similar sources |
| Estimate | Estimated by interpolation or model |
| Not obtained | Not yet obtained and not used in scores |
| Old vintage | Value before revision |
- We do not overwrite data. Revisions are appended as new vintages. Prior-month scores do not change.
- Market data use month-end values, and commodities and rates use monthly averages, as a rule.
- We do not decide scores before seeing raw data. We check drift over the last three months every month.
- L moves only on a step of +0.5 or more (from September 2026).
- Country SURI values are model values based on public information. They are not official assessments or ratings of any country.
Limits and next version
- M and L are expert judgments. Retroactive scores are not official index values of the time.
- Financial indicators are US-centered. Fed lending covers H.4.1 loans only, without asset purchases or swap lines.
- The FAO real index, grain stocks-to-use, the count of export-limiting countries, Food Gap, Rhine water levels and the OFR index have not been obtained.
- proxy-GRCS has no climate or AI axis.
- In the next report we review the scoring criteria because the gap between judgment and objective measures exceeded ±15. Anchors are reviewed in the annual validation each January.