Backtesting Results

Tested Against Real History

We backtested the Cascade algorithm against 9 real-world crises spanning geopolitical shocks, financial meltdowns, pandemics, and commodity wars. 121 entity predictions. No cherry-picking.

91%
Direction Accuracy
110 / 121 correct
86%
Acceptable Grade (A+B)
104 / 121 acceptable
9
Historical Crises
Historical crises
9 / 9
Systems Stable
Systems stable

How we grade predictions

Each entity prediction is graded on both direction (did it go the right way?) and magnitude (was the size of the move reasonable?).

A
Correct direction, magnitude within 0.5–2x of actual
B
Correct direction, magnitude within 0.25–4x of actual
C
Correct direction, magnitude off by more than 4x
F
Wrong direction

9 crises, 121 predictions

Click through each benchmark to see every entity prediction, the actual market outcome, and our grade. We show our failures alongside our successes.

01
Russia-Ukraine Invasion
Feb 24, 2022 · Geopolitical crisis + energy shock + sanctions
92%Direction
92%A+B
8.7%MAE

The model correctly captures energy price spikes, commodity disruption, and differentiated impacts on Russian vs. European vs. US equities.

Russia: -41% model vs -50% actual
Natural Gas: +30% model vs +50% actual
DAX nearly perfect: -6.0% vs -10.0%
EntityModelActualGrade
Crude Oil+18.0%+25.0%A
Natural Gas+30.0%+50.0%A
Wheat+22.0%+40.0%A
Gold+5.0%+4.0%A
S&P 500-8.2%-4.5%A
NASDAQ-6.2%-5.0%A
FTSE 100-7.1%-4.0%A
DAX-6.0%-10.0%A
VIX+35.0%+57.0%A
Defense+19.1%+25.0%A
Russia-41.0%-50.0%A
Ukraine-35.2%-40.0%A
A: 10B: 1C: 0F: 1
02
COVID-19 Market Crash
Feb 20, 2020 · Pandemic + global market crash + stimulus response
93%Direction
73%A+B
50.7%MAE

COVID was a 3-sigma event with unprecedented magnitudes (VIX +486%, oil -63%). The model correctly predicts direction for 93% of entities but underestimates peak magnitudes — expected for a once-in-a-century pandemic.

US economy: -9.9% model vs -8.0% actual
Pharma correctly identified as beneficiary: +12% vs +18%
Bitcoin: -25.7% model vs -50% actual
EntityModelActualGrade
S&P 500-10.5%-34.0%B
NASDAQ+10.6%-30.0%F
FTSE 100-8.9%-33.0%B
DAX-15.9%-38.0%B
Nikkei-5.5%-30.0%C
VIX+42.5%+486.0%C
Crude Oil-29.8%-63.0%B
Gold-4.7%+1.0%C
US 10Y-17.8%-69.0%B
Bitcoin-25.7%-50.0%A
Travel-29.2%-65.0%B
Hospitality-26.0%-55.0%B
Pharma+12.0%+18.0%A
China-15.0%-10.0%A
US-9.9%-8.0%A
A: 5B: 6C: 3F: 1
03
Brexit Vote
Jun 23, 2016 · Political shock — regional, contained
92%Direction
92%A+B
3.5%MAE

Best-performing benchmark by grade rate. The model correctly isolates Brexit as a regional UK/EU shock with limited US impact. Five entities predicted within 1% of actual.

GBP: -14.1% model vs -12.0% actual
DAX nearly perfect: -6.9% vs -6.8%
FTSE 250 hit harder than FTSE 100 (correctly)
EntityModelActualGrade
GBP-14.1%-12.0%A
UK Banking-17.7%-20.0%A
FTSE 100-4.4%-5.6%A
FTSE 250-10.0%-12.0%A
DAX-6.9%-6.8%A
Euro Stoxx-8.0%-8.6%A
S&P 500+6.1%-3.6%F
NASDAQ-4.2%-4.1%A
Nikkei-7.0%-7.9%A
Gold+5.0%+4.7%A
VIX+30.0%+49.0%A
US 10Y-8.0%-12.0%A
A: 11B: 0C: 0F: 1
04
US-China Trade War
Mar–Jun 2018 · Tariff escalation — slow-burn, bilateral
92%Direction
92%A+B
8.5%MAE

The model handles slow-burn escalation well, correctly propagating tariff impacts through commodity and equity channels.

Soybeans: -13.0% model vs -15.0% actual
Steel prices: +12.2% model vs +25.0% actual
Shanghai/Hang Seng hit harder than US indices (correctly)
EntityModelActualGrade
S&P 500-5.1%-10.0%A
NASDAQ-5.1%-8.0%A
Shanghai-7.6%-15.0%A
Hang Seng-9.7%-12.0%A
DAX-5.9%-7.0%A
VIX+28.3%+80.0%B
Soybeans-13.0%-15.0%A
Steel Price+12.2%+25.0%B
Copper-5.7%-10.0%A
Gold+3.2%-2.0%F
US Agriculture-12.8%-8.0%A
China-5.6%-3.0%A
A: 9B: 2C: 0F: 1
05
Lehman Brothers Collapse
Sep 15, 2008 · Financial crisis — banking contagion, credit freeze
87%Direction
73%A+B
30.5%MAE

Like COVID, Lehman was a multi-sigma systemic event. The model correctly identifies direction for 87% of entities but underestimates peak magnitudes in a credit-driven cascade.

Gold near-perfect: +14.5% model vs +15.1% actual
US economy: -5.8% model vs -8.0% actual
Credit spreads correctly widened: +19.8% vs +65.0%
EntityModelActualGrade
S&P 500-12.9%-28.5%B
NASDAQ-5.9%-25.2%C
FTSE 100-8.6%-24.3%B
DAX-8.1%-28.0%B
Nikkei-9.3%-34.3%B
VIX+23.5%+208.0%C
Gold+14.5%+15.1%A
Crude Oil-13.1%-46.0%B
US 10Y-5.2%+17.0%F
US Banking-15.0%-49.0%B
EU Banking-12.5%-42.0%B
USD/EUR-2.2%+6.8%F
Credit Spreads+19.8%+65.0%B
US-5.8%-8.0%A
Eurozone-7.5%-6.0%A
A: 3B: 8C: 2F: 2
06
Fukushima Disaster
Mar 11, 2011 · Natural disaster + nuclear crisis — regional with sector contagion
100%Direction
100%A+B
3.4%MAE

Best-performing benchmark across all 9 events. 100% direction accuracy with 3.4% MAE. The model excels at regional shocks with clear sectoral transmission channels.

Nuclear sector: -35.8% model vs -35.0% actual
TOPIX within 0.3%: -17.3% vs -17.0%
Renewable energy beneficiary correctly identified
EntityModelActualGrade
Nikkei-14.7%-17.5%A
TOPIX-17.3%-17.0%A
S&P 500-4.1%-3.6%A
DAX-6.5%-6.3%A
VIX+25.3%+51.5%B
Gold+2.8%+1.9%A
Crude Oil-6.7%-4.0%A
Uranium-23.3%-25.0%A
Nuclear Sector-35.8%-35.0%A
JP Insurance-14.8%-13.0%A
Renewable Energy+7.4%+12.0%A
JPY+6.9%+7.8%A
US 10Y-6.8%-7.6%A
A: 12B: 1C: 0F: 0
07
China Stock Crash / Yuan Devaluation
Aug 11, 2015 · Emerging market shock — currency devaluation, global growth fears
93%Direction
86%A+B
22.6%MAE

Strong performance on an EM-originating crisis. The model correctly propagates the China shock through commodity channels and into developed markets.

S&P 500 near-perfect: -11.0% model vs -11.2% actual
Crude Oil near-perfect: -11.1% vs -11.4%
AUD correctly fell on China exposure: -5.3% vs -4.5%
EntityModelActualGrade
Shanghai-13.1%-21.8%A
Hang Seng-8.8%-15.6%A
S&P 500-11.0%-11.2%A
NASDAQ-7.4%-13.3%A
DAX-10.2%-18.3%A
Nikkei-10.9%-14.6%A
VIX+55.7%+316.0%C
Crude Oil-11.1%-11.4%A
Copper+5.1%-6.0%F
Gold+4.7%+6.0%A
EEM-7.5%-11.4%A
CNY-4.3%-2.9%A
AUD-5.3%-4.5%A
US 10Y-7.3%-11.6%A
A: 12B: 0C: 1F: 1
08
Saudi-Russia Oil Price War
Mar 8, 2020 · Commodity shock — supply glut, oil producer conflict
85%Direction
85%A+B
16.8%MAE

Good performance on a commodity-driven crisis. Oil prices well-captured, and petrocurrency contagion works well. Gold correctly modeled as falling in a liquidity crisis — a nuanced call.

WTI: -27.7% model vs -49.0% actual
Ruble correctly fell: -10.7% vs -17.2%
Gold correctly fell (liquidity crisis): -8.8% vs -11.7%
EntityModelActualGrade
WTI-27.7%-49.0%A
Brent-27.4%-45.0%A
Natural Gas+5.3%-7.7%F
S&P 500-11.0%-7.6%A
XLE-18.3%-39.1%B
Aramco-8.6%-25.0%B
MOEX+11.4%-21.5%F
Ruble-10.7%-17.2%A
NOK-7.6%-20.1%B
CAD-5.2%-7.4%A
VIX+29.0%+97.4%B
Gold-8.8%-11.7%A
US High Yield-14.3%-15.0%A
A: 7B: 4C: 0F: 2
09
UK Gilt Crisis / Liz Truss Mini-Budget
Sep 23, 2022 · Fiscal policy shock — sovereign credibility crisis
91%Direction
91%A+B
6.3%MAE

The model handles this fiscal policy shock well — a crisis type not seen in the original calibration set. GBP near-perfect, and limited global spillover correctly modeled.

GBP: -9.8% model vs -8.1% actual
FTSE 250 hit harder than FTSE 100 (correctly)
UK LDI pension fund crisis captured: -22.8% vs -40.0%
EntityModelActualGrade
GBP-9.8%-8.1%A
UK Gilts 30Y+14.0%+35.3%B
FTSE 100-5.8%-4.8%A
FTSE 250-9.1%-8.5%A
UK Banking-8.4%-10.0%A
S&P 500-3.5%-1.7%B
VIX+4.7%+8.0%A
Gold-3.0%-2.9%A
EUR/GBP-7.9%+5.4%F
UK LDI-22.8%-40.0%A
UK-12.4%-5.0%B
A: 7B: 3C: 0F: 1

Performance by crisis type

The model performs best on regional, contained shocks and worst on systemic financial crises where magnitudes exceed calibration range.

Crisis TypeBenchmarksAvg A+B
GeopoliticalRussia-Ukraine, Trade War92%
Financial / SystemicLehman, COVID73%
Political / RegionalBrexit, UK Gilt Crisis92%
Natural DisasterFukushima100%
CommodityOil Price War85%
EM / CurrencyChina Crash86%

Known limitations

No model is perfect. We believe in showing where we fall short so you know exactly what you're working with.

Systemic financial crises
Both Lehman (73%) and COVID (73%) underperform. The model is calibrated for 1-2 sigma shocks, not credit-driven cascading failures with 3+ sigma magnitudes.
VIX magnitude
Consistently underestimated. The model captures VIX direction correctly but peak magnitudes are 2-10x too low due to VIX's convex, nonlinear behavior.
Currency cross-rates
EUR/GBP and USD/EUR are direction misses. Cross-rates depend on relative dynamics between two economies — the model treats each currency independently.
Recovery timing
The model shows sustained drawdowns correctly, but exact recovery day predictions remain noisy due to stochastic perturbations.

How the model works

A high-level overview of the Cascade propagation engine. Specific equations and parameters are proprietary.

01
Directed Graph
Entities (countries, sectors, markets, commodities) are nodes in a weighted, directed graph. Edges represent transmission channels with configurable delays and nonlinear transfer functions.
02
4D Vector State
Each entity carries a 4-dimensional state vector — Stability, Economic, Political, and Social — enabling asymmetric responses across different domains of impact.
03
Discrete-Time Cascade
At each time step, shocks propagate through the network with entity-specific persistence, sensitivity, and restoration dynamics. The system is guaranteed stable via spectral radius analysis.
04
Monte Carlo Projection
Hundreds of independent simulation runs with stochastic noise produce confidence intervals and probability distributions for recovery timelines and worst-case scenarios.
05
Self-Calibration
A continuous calibration engine compares model predictions against observed market data, automatically proposing and applying parameter adjustments when confidence thresholds are met.
06
Live Data Ingestion
RSS feeds, stock market data, and weather APIs are ingested in real-time. Events are classified, entities are extracted, and the network automatically updates with new shocks and connections.

See Cascade in action

Schedule a walkthrough to see how the algorithm performs on live, real-time data.