ADI Backtest: 83 Quarters of Validation (2005–2025)

Why a Backtest Matters

A composite index is only as credible as its behavior during the events it claims to measure. If the American Distress Index can’t identify the worst financial crisis since the Great Depression, the methodology is broken.

The ADI was not fitted to the GFC. The component weights come from principal component analysis across 42 indicators. The baseline period for Z-score normalization is 2015–2024 — entirely post-crisis. The backtest computes “what would the ADI have read during the GFC” using a methodology designed without reference to GFC-era outcomes.

This is the single most important validation test for the index.

Summary: What the Backtest Shows

Across 83 quarters from 2005 through 2025, the ADI:

• Entered Crisis (above 80) in Q3 2008 — the quarter Lehman Brothers collapsed
• Peaked at 81.2 in Q4 2009 — correctly identifying the delayed peak of household distress
• Remained in Serious Stress (65–80) through Q4 2011 — matching the slow, painful recovery
• Transitioned to Elevated (50–65) in Q1 2012 — as delinquency rates finally began normalizing
• Registered the COVID labor shock at 66.0 in Q1 2020, then dropped to 23.6 by Q1 2021 as stimulus flooded the system
• Returned to Elevated at 50.2 in Q2 2024 — where it remains today at 56.8

The GFC, Quarter by Quarter

The table below shows every quarter from 2005 through 2012. The composite score, zone, and each component’s Z-score are computed from the production ADI methodology — the same code that generates today’s reading.

Quarter	Score	Zone	Buffer Z	Debt Z	Financial Z	Cost Z	Labor Z
2005-Q1	68.2	Serious	1.06	0.38	1.16	0.53	0.05
2005-Q2	67.7	Serious	1.06	0.38	1.16	0.42	0.03
2005-Q3	68.0	Serious	1.06	0.38	1.16	0.42	0.10
2005-Q4	67.8	Serious	1.06	0.38	1.16	0.44	0.05
2006-Q1	67.4	Serious	1.06	0.38	1.16	0.50	−0.12
2006-Q2	67.9	Serious	1.06	0.38	1.16	0.54	−0.03
2006-Q3	67.7	Serious	1.06	0.39	1.16	0.46	−0.02
2006-Q4	67.0	Serious	1.06	0.47	1.16	0.12	0.01
2007-Q1	67.5	Serious	1.06	0.54	1.16	0.18	−0.02
2007-Q2	68.5	Serious	1.06	0.63	1.16	0.26	−0.03
2007-Q3	69.5	Serious	1.06	0.81	1.16	0.20	−0.02
2007-Q4	72.9	Serious	1.06	0.98	1.16	0.64	0.09
2008-Q1	74.7	Serious	1.06	1.23	1.16	0.62	0.15
2008-Q2	75.4	Serious	0.92	1.52	1.16	0.50	0.25
2008-Q3	80.6	Crisis	1.06	1.92	1.16	0.53	0.56
2008-Q4	77.9	Serious	0.75	1.98	1.16	−0.06	1.00
2009-Q1	79.7	Serious	0.73	1.98	1.16	−0.06	1.50
2009-Q2	78.4	Serious	0.53	1.98	1.16	0.02	1.46
2009-Q3	79.6	Serious	0.83	1.98	1.16	0.02	1.19
2009-Q4	81.2	Crisis	0.79	1.98	1.16	0.77	0.92
2010-Q1	78.5	Serious	0.78	1.98	0.53	0.87	0.79
2010-Q2	73.7	Serious	0.64	1.98	−0.24	0.81	0.71
2010-Q3	71.7	Serious	0.65	1.98	−0.58	0.63	0.71
2010-Q4	69.9	Serious	0.68	1.98	−1.01	0.71	0.56
2011-Q1	67.9	Serious	0.57	1.98	−1.18	0.69	0.46
2011-Q2	68.5	Serious	0.61	1.98	−1.28	0.84	0.50
2011-Q3	67.4	Serious	0.57	1.98	−1.49	0.89	0.47
2011-Q4	66.0	Serious	0.56	1.98	−1.72	0.91	0.35
2012-Q1	63.1	Elevated	0.42	1.85	−1.80	0.87	0.25
2012-Q2	61.5	Elevated	0.34	1.68	−1.64	0.68	0.29
2012-Q3	62.5	Elevated	0.51	1.57	−1.59	0.77	0.28
2012-Q4	57.3	Elevated	0.08	1.44	−1.64	0.62	0.27

Bold values indicate Z-scores above 0.5 (elevated) or below −0.5 (suppressed). The pattern tells a clear story.

Five Findings from the GFC Backtest

1. Buffer Depletion Was the First Signal

Buffer Depletion Z sat at 1.06 — maximum distress — from Q1 2005 through Q1 2008. The personal savings rate was at historic lows. Debt service ratios were elevated. Households had no margin for error.

Meanwhile, Debt Stress Z was only 0.38 in early 2005. The gap between buffers running out and payments going delinquent spanned nine quarters — matching the cross-correlation analysis documented in What the Savings Rate Told Us Nine Quarters Before the Last Crisis.

This is why Buffer Depletion carries 30% of the ADI’s weight. It’s not an editorial choice — it’s the component that warns first.

2. The Crisis Arrived Exactly When Expected

Debt Stress Z crossed 0.5 in Q2 2007, nine quarters after Buffer Depletion’s elevation began. The composite score crossed into Crisis (above 80) in Q3 2008 — the quarter of the Lehman bankruptcy. The ADI didn’t need hindsight to flag the crisis. The methodology, applied mechanically, identifies Q3 2008 as the inflection point.

3. Debt Stress Peaked and Stayed Elevated

Debt Stress Z hit the winsorized ceiling of 1.98 in Q4 2008 and stayed there through Q1 2011 — ten consecutive quarters at maximum distress. This reflects the reality on the ground: mortgage delinquency rose from 4.36% in Q2 2008 to 11.49% in Q1 2010 and remained above 10% for years. The delinquency crisis was not a spike — it was a plateau.

4. Components Rotated During Recovery

The recovery pattern reveals something the composite score alone can’t show. Financial Conditions Z turned sharply negative (credit easing) starting in Q2 2010, pulling the composite down. But Debt Stress Z remained at its ceiling. Labor Market Z gradually declined from 1.50 in Q1 2009 to 0.25 in Q1 2012.

Each component recovered at a different speed:

• Financial Conditions recovered first (credit markets reopened)
• Labor Market recovered second (unemployment slowly fell)
• Buffer Depletion recovered third (savings rates normalized)
• Debt Stress recovered last (delinquency took years to clear)

This differential recovery is exactly what a well-constructed composite should capture. A single-indicator index would miss the handoff between components.

5. The ADI Was Already Elevated Before the Crisis

A critical finding: the ADI read 67–68 (Serious Stress) throughout 2005–2006, well before any crisis was apparent. This was driven by Buffer Depletion and Financial Conditions (tight leverage). The conventional wisdom at the time was that the economy was healthy. The ADI disagrees — and the data supports the ADI.

This pre-crisis elevation is not a false positive. It reflects genuine underlying fragility that the housing boom was masking. When the mask came off, the score moved from Serious to Crisis in three quarters.

The COVID Sequence

The pandemic offers a contrasting validation. COVID was an exogenous labor market shock, not a building financial crisis. The ADI’s behavior should differ from the GFC — and it does.

Quarter	Score	Zone	Buffer Z	Debt Z	Financial Z	Cost Z	Labor Z
2019-Q4	55.2	Elevated	0.48	−0.01	0.45	0.37	−0.48
2020-Q1	66.0	Serious	−0.00	0.09	0.60	0.48	2.72
2020-Q2	39.5	Normal	−2.44	−0.11	−0.32	0.02	2.76
2020-Q3	30.5	Healthy	−2.24	−0.52	−2.61	0.36	2.76
2020-Q4	33.8	Healthy	−1.49	−0.42	−2.61	−0.07	2.37
2021-Q1	23.6	Healthy	−2.44	−0.71	−2.60	−0.02	2.18
2021-Q2	25.1	Healthy	−1.60	−1.14	−2.16	0.26	0.84
2021-Q3	26.9	Healthy	−1.16	−1.22	−1.46	−0.11	0.21
2021-Q4	31.2	Healthy	−0.67	−1.22	−1.11	0.19	−0.35

Three features distinguish the COVID pattern from the GFC:

1. Labor spiked instantly, alone. In Q1 2020, only Labor Market Z was elevated (2.72). Every other component was near baseline. During the GFC, multiple components rose together over quarters. COVID was a single-component shock.

2. Stimulus created a false “Healthy.” By Q3 2020, the ADI fell to 30.5 — Healthy — despite labor markets still being severely disrupted. The reason: stimulus checks drove Buffer Depletion Z to −2.44 (maximum savings) and emergency credit programs drove Financial Conditions Z to −2.61 (maximum easing). These two components mathematically overwhelmed the labor signal.

3. The recovery was faster but left a different scar. The ADI returned to the 50s by Q2 2024 — but driven by Buffer Depletion (savings being spent down) and Debt Stress (delinquency normalizing), not by the labor shock that started the cycle. The distress rotated from labor market disruption to household balance sheet erosion.

Component Contribution Analysis

The ADI’s five-component structure means each component’s contribution (Z-score × weight × 27) can be decomposed. At key moments:

Q3 2008 (Crisis entry, score 80.6):

• Buffer Depletion: +8.6 points (35% of excess above 50)
• Debt Stress: +13.0 points (42%)
• Financial Conditions: +4.7 points (15%)
• Cost Pressure: +2.2 points (7%)
• Labor Market: +2.3 points (7%)

Debt Stress was the largest contributor at the moment of crisis entry. But Buffer Depletion had been contributing +8.6 points since 2005 — three years of sustained warning.

Q4 2009 (Peak, score 81.2):

• Buffer Depletion: +6.4 points (20%)
• Debt Stress: +13.3 points (43%)
• Financial Conditions: +4.7 points (15%)
• Cost Pressure: +3.1 points (10%)
• Labor Market: +3.7 points (12%)

At the peak, all five components were contributing positively. This broad-based signal is the hallmark of a genuine crisis — not a single-indicator spike.

Q1 2021 (Stimulus floor, score 23.6):

• Buffer Depletion: −19.8 points
• Debt Stress: −4.8 points
• Financial Conditions: −10.5 points
• Cost Pressure: −0.1 points
• Labor Market: +8.8 points

Stimulus overwhelmed the labor signal by 26 points on the downside. The composite correctly reflected that households were, on average, better buffered than normal — even if the labor market was in distress.

False Positive Analysis

A valid concern: does the ADI produce false positives — elevated readings without a subsequent crisis? Three periods warrant examination:

2005–2006 (67–68, Serious Stress): The ADI read Serious Stress for two years before the crisis materialized. This was not a false positive. It was an early warning that the conventional narrative (“everything is fine”) missed. The Buffer Depletion component was correctly identifying fragility that banking statistics hadn’t yet reflected.

2015–2019 (53–60, Elevated): The ADI remained in the Elevated zone throughout the longest expansion in U.S. history. Is this a false positive? We argue no — the post-GFC recovery left household balance sheets structurally weaker than the pre-2000 norm. Savings rates were lower, debt service was higher, and financial conditions were tighter than the pre-Great Moderation baseline. “Elevated” doesn’t mean “crisis is imminent.” It means household buffers are thinner than the historical norm.

2022–2023 (42–48, Normal): After the stimulus-driven Healthy readings of 2021, the ADI returned to Normal. This period had no crisis — correctly. The ADI fell from Healthy through Normal as stimulus savings were depleted, then climbed back toward Elevated as underlying distress signals strengthened.

The ADI has never read Crisis without a recession in progress. Its Serious Stress readings have always occurred during periods of genuine household vulnerability — even if the crisis hadn’t yet manifested in headline data.

Comparison to Other Indices

The Chicago Fed National Financial Conditions Index (NFCI) is the most widely cited financial stress measure. It spiked during the GFC (peaking in Q4 2008) and normalized rapidly by mid-2009. The NFCI tracks financial market conditions — it does not measure household distress directly.

The ADI diverges from the NFCI in two important ways:

1. Earlier warning. The NFCI was near zero (average conditions) through 2006–2007. The ADI was at 67–68 (Serious Stress). The ADI saw household fragility that financial market indicators missed.

2. Slower normalization. The NFCI returned to normal by mid-2009. The ADI didn’t reach Elevated until Q1 2012 — three years later. This reflects the difference between financial markets (which recovered when the Fed intervened) and household balance sheets (which took years to repair). For measuring actual household distress, the ADI’s slow normalization is more accurate.

The 9-Quarter Lag Evidence

The backtest provides the clearest evidence for the ADI’s most consequential finding: that Buffer Depletion leads Debt Stress by approximately 9 quarters.

In the data:

• Buffer Depletion Z crossed above 1.0 in Q1 2005
• Debt Stress Z crossed above 0.5 in Q2 2007 (9 quarters later)
• Debt Stress Z crossed above 1.0 in Q1 2008 (12 quarters after Buffer’s peak)
• Debt Stress Z hit its ceiling in Q4 2008 (15 quarters after Buffer’s peak)

The lag is not exact — it ranges from 9 to 15 quarters depending on the threshold chosen. But the direction is unambiguous: savings collapse comes first, delinquency follows. This is documented with cross-correlation analysis at r = 0.69 in the original leading indicator paper and confirmed by the systematic scanner across 57,541 indicator pairs.

Current Reading in Historical Context

As of Q3 2025, the ADI reads 56.8 — Elevated. How does this compare to the backtest periods?

Period	ADI Range	Zone	Key Driver
Pre-GFC build (2005–2006)	67–68	Serious	Buffer + Financial Conditions
GFC crisis (2008-Q3 to 2009-Q4)	78–81	Serious/Crisis	Debt Stress + Labor
Post-GFC recovery (2012–2014)	57–63	Elevated	Debt Stress (lingering)
Late expansion (2018–2019)	53–59	Elevated	Financial Conditions
COVID floor (2021-Q1)	23.6	Healthy	Stimulus + easing
Current (2025-Q3)	56.8	Elevated	Buffer + Debt + Financial

The current reading (56.8) is structurally similar to the 2013–2014 post-GFC recovery period — the ADI at a comparable level but with different component mix. In 2013, Debt Stress was the dominant contributor (still elevated from the GFC). Today, the signal is more evenly distributed across Buffer Depletion, Debt Stress, and Financial Conditions.

This broader base is worth watching. The GFC backtest shows that the most dangerous configurations are when multiple components are elevated simultaneously. A single-component spike (like COVID’s labor shock) produces a sharp but short distress reading. Broad-based elevation — like 2005–2007 — produces sustained, escalating distress.

Calibration Assessment

The zone thresholds — Healthy (below 35), Normal (35–50), Elevated (50–65), Serious (65–80), Crisis (above 80) — were calibrated against subjective assessments of economic conditions across the backtest period:

Zone	Backtest Quarters	Real-World Conditions
Crisis (above 80)	2 quarters (2008-Q3, 2009-Q4)	Lehman collapse, peak unemployment, foreclosure wave
Serious (65–80)	26 quarters (2005–2011)	Pre-crisis fragility → GFC → slow recovery
Elevated (50–65)	31 quarters (2012–2019, 2024–2025)	Post-GFC normalization, late expansion, current
Normal (35–50)	11 quarters (2019–2020, 2022–2023)	Pre-COVID stability, post-stimulus normalization
Healthy (below 35)	7 quarters (2020-Q3 to 2021-Q4)	Stimulus-driven artificial surplus

Crisis is rare (2 of 83 quarters, 2.4%) and corresponds to the most severe household distress in modern history. Healthy is also rare (7 quarters, 8.4%) and corresponds to extraordinary fiscal intervention. The distribution is consistent with a well-calibrated index — extreme readings are extreme.

Methodology Notes

Baseline period: Z-scores are computed against Q1 2015 through Q4 2024. This baseline is entirely post-GFC, so the backtest is genuinely out-of-sample — the GFC-era data was not used to set the normalization parameters.

Winsorization: Component values are capped at the 95th percentile before computing baseline statistics. This prevents COVID-era outliers (savings rate spiking to 33.8% in Q2 2020) from distorting the baseline standard deviation. Without winsorization, COVID would compress the GFC signal.

Z-anchored scaling: Each quarter’s score depends only on its own Z-score relative to the baseline. No quarter can distort another quarter’s score. This means adding new data (future quarters) cannot retroactively change historical scores — the backtest is stable.

Reproducibility: The entire backtest can be reproduced by running python3 scripts/compute_adi.py. The output matches this article’s tables exactly because both use the same code path.

Data Access

The full quarterly ADI dataset — all 83 quarters with composite scores, zone assignments, and component Z-scores — is available via the API:

• JSON: /api/adi.json
• Methodology: ADI Methodology
• Leading indicators: Structural Projections Methodology
• Current outlook: /outlook
• Print summary: ADI One-Pager (current score, components, key statistics)

Source data: Federal Reserve Economic Data (FRED), Bureau of Labor Statistics (BLS), Chicago Fed NFCI. All underlying series are publicly available. FRED attribution: Federal Reserve Bank of St. Louis. For current default rates across all loan types, see our Default Rates Statistics roundup.