Market Stress Monitor
Asset Management Risk Dashboard
Data: FRED + Yahoo Finance
Refresh: Daily EOD
Dashboard
Indicators
Backtest
Models
Forecast
Market Stress Level
24.1 +12.0 (5d)
CALM
Risk-on conditions
As of 2026-01-16 | Updated 12:24
HIGH: Safe Haven (GLD/SPY) at 99th percentile
Signal Breakdown (by contribution)
! Safe Haven (GLD/SPY) 6% wt 99
VIX Term Structure 8% wt 77
VIX 15% wt 25
Sector Correlation 8% wt 23
USD Index 8% wt 22
Defensive Rotation 10% wt 22
HY Credit Spread 20% wt 7
IG Credit Spread 15% wt 6
S&P 500 Drawdown 10% wt 1
Current Levels
HY Credit Spread 2.71%
VIX 15.8
IG Credit Spread 0.76%
USD Index 120.6
Defensive Rotation 0.355
Sector Correlation 0.187
VIX Term Structure 0.835 (backwardation)
Safe Haven (GLD/SPY) 0.609
S&P 500 Drawdown 6940

Indicator Reference Guide

Each indicator in the composite score is selected based on backtesting against historical stress events.

High Yield Credit Spread

FRED (BAMLH0A0HYM2)
Low: 7
Weight: 20%

The spread between high yield (junk) bonds and treasuries. Widens sharply during stress as investors demand more compensation for risk.

Interpretation
Higher = more stress. During crises, spreads can exceed 10%. Normal range is 3-5%.
Backtest Performance
Signal Ratio: 2.20x | Detection: 100% (7/7 events)

Investment Grade Credit Spread

FRED (BAMLC0A0CM)
Low: 6
Weight: 15%

Option-adjusted spread for investment grade corporate bonds vs treasuries. Covers the full IG spectrum including BBB (VIF analysis showed BBB was redundant).

Interpretation
Higher = more stress. IG spreads widen when investors flee corporate bonds for safer assets.
Backtest Performance
Signal Ratio: 2.15x | Detection: 100% (7/7 events)

VIX (Volatility Index)

FRED (VIXCLS)
Low: 25
Weight: 15%

The 'fear gauge' - measures expected 30-day S&P 500 volatility implied by options prices.

Interpretation
Higher = more fear. Above 30 is elevated, above 40 is panic.
Backtest Performance
Signal Ratio: 1.90x | Detection: 100% (7/7 events)

Defensive Rotation (XLU/XLY)

Yahoo Finance (derived)
Low: 22
Weight: 10%

Ratio of Utilities (defensive) to Consumer Discretionary (cyclical). Rising ratio indicates rotation to safety.

Interpretation
Higher = more defensive positioning. Spikes during risk-off periods.
Backtest Performance
Signal Ratio: 1.90x | Detection: 100% (2/2 events)

USD Index

FRED (DTWEXBGS)
Low: 22
Weight: 8%

Trade-weighted US dollar index. Dollar strengthens as global reserve currency during crises.

Interpretation
Higher = flight to safety. Strong dollar can also create EM stress.
Backtest Performance
Signal Ratio: 1.72x | Detection: 100% (7/7 events)

Sector Correlation

Yahoo Finance (derived)
Low: 23
Weight: 8%

Average pairwise correlation of S&P sector ETFs (60-day rolling). High correlation means diversification breaks down.

Interpretation
Higher = more stress. When all sectors move together, it signals panic selling.
Backtest Performance
Signal Ratio: 1.76x | Detection: 100% (2/2 events)

VIX Term Structure

Yahoo Finance (VIX/VIX3M)
Elevated: 77
Weight: 8%

Ratio of spot VIX to 3-month VIX. Normally in contango (>1). Backwardation (<1) signals panic.

Interpretation
Below 1.0 = backwardation = panic. Below 0.9 = severe stress.
Backtest Performance
Signal Ratio: 1.67x | Detection: 100% (2/2 events)

Safe Haven Demand (GLD/SPY)

Yahoo Finance (derived)
Elevated: 99
Weight: 6%

Ratio of gold to S&P 500. Rising ratio indicates flight to gold as safe haven.

Interpretation
Higher = more fear. Gold outperforms equities during stress.
Backtest Performance
Signal Ratio: 1.42x | Detection: 100% (2/2 events)

S&P 500 Drawdown

FRED (SP500)
Low: 1
Weight: 10%

Current drawdown from 52-week high. Direct measure of equity market stress.

Interpretation
Higher = deeper drawdown. 10%+ is correction, 20%+ is bear market.
Backtest Performance
Signal Ratio: N/A | Detection: 100%

Backtesting Methodology & Results

How We Test Indicators

Each indicator is evaluated against 6 major market stress events:

  • 2008-09: Financial Crisis (56.8% drawdown)
  • 2011: Debt Ceiling Crisis (19.4% drawdown)
  • 2015: China Devaluation (12.4% drawdown)
  • 2018: Fed Tightening (19.8% drawdown)
  • 2020: COVID Crash (33.9% drawdown)
  • 2022: Rate Shock (25.4% drawdown)

Signal Ratio

Average percentile during stress / Average during normal times.

Higher is better. 2.0x = twice as elevated during stress.

Detection Rate

% of stress events where indicator exceeded 70th percentile.

Higher is better. 100% = caught every major event.

Indicator Rankings

IndicatorSignal RatioDetectionTier
HY Credit Spread2.19x100%1
BBB Credit Spread2.18x100%1
VIX1.89x100%2
Defensive Rotation1.89x100%2
USD Index1.73x100%2
Sector Correlation1.75x100%3
VIX Term Structure1.63x100%3
Safe Haven (GLD/SPY)1.38x100%4
S&P 500 DrawdownN/A100%4

What Didn't Work

IndicatorSignal RatioProblem
RSI (SPY, QQQ, IWM)0.53-0.58xGoes DOWN during stress
10Y-2Y Spread0.61xOnly 33% detection - too slow
SKEW Index0.22xOpposite of expected
Fed Funds Rate1.02xPolicy-driven, not market

Linear vs Non-Linear Models

Testing whether non-linear relationships improve fast market detection.

What Linear Models Miss

  • Convexity at extremes - Going from 70th to 85th percentile is fundamentally different than 40th to 55th.
  • Convergence effects - When VIX, credit, and USD all spike together, it's multiplicative, not additive.
  • Regime shifts - In calm markets, credit leads. In stress, everything correlates.
  • Lead-lag relationships - Credit spreads widen before equities fall.

Backtest Results: 6 Major Stress Events

EventLinearHybrid
Financial Crisis (2008)7295
Debt Ceiling (2011)7295
China Devaluation (2015)7295
Fed Tightening (2018)7193
COVID Crash (2020)7395
Rate Shock (2022)7299

Model Performance

Signal Ratio
+24%
Hybrid vs Linear
Peak Score
95-99
Hybrid in stress
Lead Time
+0.8d
Early warning
Precision
-28%
More false positives

Practical Recommendation

Two-Stage Monitoring Approach

STAGE 1 Use Linear model for daily monitoring. Fewer false alarms, stable baseline.
STAGE 2 When Linear crosses 50+, switch to Hybrid lens for amplified signal.
STAGE 3 If Hybrid exceeds 85, treat as confirmed fast market.

Large Move Forecaster

Neural network model to predict S&P 500 moves greater than 3% (either direction) using t-1 (previous day) indicator values.

Current Forecast

Probability of Large Move
0.0%
NORMAL
Model Prediction
Normal Day Expected
Threshold: 25%
Actual Outcome
Normal (+0.65%)
Date: 2026-01-09

Model Performance (Hold-Out Test Set)

Accuracy
97.8%
Precision
50.0%
Recall
20.0%
AUC-ROC
0.876
Training Data: 925 samples (12 large moves)
Test Data: 232 samples (5 large moves)

Walk-Forward Backtest (2020-Present)

Detection Performance

Recall (Sensitivity) 75.0%
Large Moves Caught 9 / 12
False Alarms 429
Precision 2.1%

Confusion Matrix

Pred: Normal Pred: Large
Actual: Normal 447 429
Actual: Large 3 9

Feature Importance (Approximate)

usd_t1
7.9%
sector_corr_t1
7.8%
vix_term_t1
7.8%
hy_spread_5d_change
7.6%
safe_haven_t1
7.4%
vix_5d_std
7.3%
sp500_20d_vol
7.3%
ig_spread_t1
7.1%
defensive_t1
7.1%
sp500_5d_return
6.8%

Methodology

Model Architecture

  • Type: Multi-Layer Perceptron (MLP) Neural Network
  • Layers: 64 - 32 - 16 neurons (3 hidden layers)
  • Activation: ReLU with adaptive learning rate
  • Regularization: L2 (alpha=0.01) + early stopping
  • Class Balancing: Oversampling (target 15% positive)

Key Design Choices

  • Target: Binary (|return| > 3%)
  • Features: t-1 lagged indicators + rolling stats
  • Prediction Threshold: 0.25 (lowered from 0.5 for rare events)
  • Validation: Walk-forward (no lookahead bias)
  • Priority: High recall (catch moves) over precision

Interpretation Guide

ProbabilitySignalSuggested Action
0-15%Normal conditionsStandard risk protocols
15-30%Slightly elevatedMonitor closely, review hedges
30-50%Elevated riskConsider reducing exposure, tighten stops
50%+High probability of large moveActive risk management, ensure liquidity

Important Notes

  • Large moves (>3%) are rare (~1% of days), so precision is inherently low
  • The model is tuned for high recall (catching moves) at the cost of false alarms
  • Use in conjunction with the composite stress score, not as a standalone signal