Backtest Analysis
A retroactive analysis of how the v3.1 framework's signals would have behaved during three major market events. This analysis is not a prediction of future performance — see the limitations section below.
What happened
Massive overinvestment in internet infrastructure. Telecom and networking companies spent billions building capacity that wouldn't be used for a decade. The Nasdaq peaked in March 2000 and fell 78% over the next 2.5 years.
How signals would have performed
Signal 1 (Capex Divergence): Would have been strongly Red. Capital expenditure across tech companies was growing at 50–100%+ annually while revenue growth was often negative or single digits. The divergence was extreme.
Signal 2 (Private Credit Stress): Limited data available for this period. Private credit markets were far less developed in 2000. This signal would have had minimal utility — a known limitation.
Signal 3 (HY Spreads): High-yield spreads widened significantly in late 2000/early 2001. Would have been Yellow-to-Red, sector-driven. However, the widening was somewhat delayed relative to the equity peak.
Signal 4 (Semiconductor Leadership): SOXX (launched 1998) significantly underperformed the broader market starting in Q3 2000. Would have flagged Red.
Signal 5 (Market Breadth): Breadth collapsed dramatically. By late 2000, the majority of tech stocks were well below their 200DMAs while a few mega-caps held up the indices. Would have been deep Red.
Signal 6 (Margin Debt): Margin debt peaked in March 2000 and declined for several consecutive months. Would have been Red by mid-2000.
Assessment
The system would likely have reached Level 2–3 by Q3-Q4 2000 — several months after the March peak but well before the worst of the decline. The structural signals (especially capex divergence) would have been flashing warning signals even earlier, possibly in 1999.
Limitation: Private credit data is essentially nonexistent for this period, so one of the six signals would have been unavailable.
What happened
A credit-driven financial crisis centered on mortgage-backed securities and banking. The S&P 500 fell 57% from peak to trough.
How signals would have performed
Signal 1 (Capex Divergence): Tech capex was not the driver of this crisis. Divergence would have been moderate — Yellow at most. This was a financial sector crisis, not a tech infrastructure crisis.
Signal 2 (Private Credit Stress): Would have been strongly Red — but this was a broad credit crisis, not sector-specific to tech. The system is designed for sector bubbles.
Signal 3 (HY Spreads): Spreads exploded to historic highs. However, the proper classification would have been Macro-Driven, which under our framework reclassifies Signal 3 as Context with zero escalation weight.
Signals 4-6 (Peak signals): All would have been Red as the entire market collapsed.
Assessment
This is the most important honest assessment: the 2008 crisis was not a tech bubble. The system would have detected the market-wide distress through peak signals but the structural layer would not have been strongly activated for the right reasons. The system would likely have reached Level 1–2, which would have provided some protection, but the alert would have been partially for the wrong reasons.
This is actually by design — the system is built to detect sector-specific bubble dynamics, not general market crashes. The sector-linkage veto would have correctly reclassified much of the credit stress as macro-driven.
What happened
A broad tech correction driven primarily by interest rate normalization. The Nasdaq fell approximately 33% from its November 2021 peak. This was a valuation correction, not a credit crisis or infrastructure bust.
How signals would have performed
Signal 1 (Capex Divergence): Moderate. Cloud capex was growing but revenue was still growing reasonably. Yellow at most.
Signal 2 (Private Credit Stress): Some stress visible in late 2022 but not severe. Green to Yellow.
Signal 3 (HY Spreads): Spreads widened but primarily macro-driven (Fed rate hikes). The sector-linkage toggle would correctly classify this as macro-driven Context.
Signal 4 (Semiconductor Leadership): SOXX underperformed significantly — down ~35% peak-to-trough vs SPY's ~25%. Would have been Red.
Signal 5 (Market Breadth): Breadth collapsed across the tech sector. Would have been Red.
Signal 6 (Margin Debt): Margin debt declined for several consecutive months starting in early 2022. Would have been Red.
Assessment
The system would have reached Level 1–2, which is appropriate for a significant correction. Critically, it would not have reached Level 3 (Cascade Confirmed) because the structural signals were not strongly activated — there was no massive capex-revenue divergence and credit stress was macro-driven, not sector-driven. This is the correct response: 2022 was a correction, not a bubble burst.
Honest Limitations
The following limitations must be understood when interpreting these backtest results:
- n=3. Three historical events is statistically insufficient for any meaningful confidence. This backtest provides directional intuition, not statistical validation.
- Hindsight bias. The framework was designed after studying these events. It is impossible to fully eliminate the circularity of designing a system that "would have worked" on the data that inspired it.
- Data gaps. Private credit data is sparse before 2015. The SOXX ETF launched in 1998. Margin debt data has always been delayed. The farther back you go, the less reliable the analysis.
- Survivorship bias in stock selection. The 14-stock breadth basket was selected in 2026 based on today's AI ecosystem. Some of these companies didn't exist or weren't relevant during earlier periods.
- Peak signals are confirmatory, not predictive. They detect that the unwind has begun, not that it will begin. By the time peak signals fire, some decline has already occurred.
- The next bubble will be different. Every market cycle has unique characteristics. The AI infrastructure cycle may end through a channel this system doesn't monitor.
Past performance — including backtested performance — is not indicative of future results.