How It Works

Bubble Trouble uses a two-horizon architecture with six signals organized across three layers. The system has one objective: detect when the AI infrastructure investment cycle has become a bubble and, critically, when that bubble is beginning to deflate.

The Two-Horizon Architecture

The framework separates detection into two distinct time horizons, each answering a different question:

Structural Horizon (Slow)

Question: Has a bubble regime formed?

This layer monitors the fundamental economics of the AI buildout. When companies are spending dramatically more on AI infrastructure than the revenue it generates, and when the financing markets that support this spending show stress, the structural conditions for a bubble are in place.

This layer moves slowly — quarters, not weeks — because structural imbalances take time to build and are durable once established.

Peak Detection Horizon (Fast)

Question: Is the unwind beginning?

This layer monitors market-level indicators that signal whether the unwind has started. Semiconductor leadership failure, breadth collapse in AI stocks, and margin debt rollover are confirmatory signals — they tell you the bubble is deflating, not that it will deflate.

This layer moves fast — weeks — because market reversals can accelerate quickly once they begin.

The Six Signals

Signal 1 — Capex vs Revenue Divergence (Structural)

Tracks the gap between how much big tech companies (Microsoft, Google, Amazon, Meta) are spending on AI infrastructure and how much revenue that spending generates. When capex growth dramatically outpaces revenue growth, it suggests overinvestment — the classic pattern in every infrastructure bubble.

Thresholds: Green if divergence is under 15 percentage points; Yellow at 15–30pp; Red above 30pp.

Update frequency: Quarterly, following earnings reports.

Signal 2 — Private Credit Stress Basket (Structural)

A composite of four sub-indicators monitoring stress in private credit markets: BDC (Business Development Company) discounts to net asset value, redemption gate incidents, bank financing pullback from private credit, and PIK (payment-in-kind) loan stress in tech credit books. Private credit is less transparent than public markets, so stress here often appears before it shows up in public bond spreads.

Thresholds: Green if 0–1 sub-indicators are stressed; Yellow if 2 are stressed; Red if 3–4 are stressed.

Update frequency: Weekly, manually assessed from public filings and financial news.

Signal 3 — HY Tech OAS Spreads (Financing)

Monitors high-yield bond option-adjusted spreads (BAMLH0A3HYC from FRED) expressed as a rolling five-year percentile. When spreads widen, it means the cost of financing is rising — a stress indicator for companies reliant on credit markets.

Critical feature: This signal has a sector-linkage toggle. If spread widening is driven by macro factors (Fed tightening, inflation) rather than sector-specific stress, it is reclassified as context information with zero escalation weight. Only sector-linked spread widening counts.

Thresholds: Green below 50th percentile; Yellow at 50th–75th; Red above 75th.

Update frequency: Weekly via FRED data.

Signal 4 — Semiconductor Leadership (Peak Detection)

Tracks the four-week rolling return of SOXX (semiconductor ETF) minus SPY (S&P 500). In a healthy AI boom, semiconductors lead the market. When they begin to underperform significantly, it signals that the market's confidence in the AI buildout is cracking.

Thresholds: Green if SOXX is outperforming or within -5% of SPY; Yellow at -5% to -10%; Red if worse than -10%.

Update frequency: Weekly via market data.

Signal 5 — Market Breadth (Peak Detection)

Measures the percentage of a 14-stock AI ecosystem basket trading above their 200-day moving average. Breadth collapse — where fewer and fewer stocks hold up the index — is a classic late-cycle signal. The basket includes: MSFT, GOOGL, AMZN, META, NVDA, AMD, AVGO, ASML, AMAT, EQIX, DLR, CRM, NOW, SNOW.

Thresholds: Green above 70%; Yellow at 50–70%; Red at 35–50%; Red (severe) below 35%.

Update frequency: Weekly via market data.

Signal 6 — Margin Debt Rollover (Peak Detection)

Monitors the trend in FINRA-reported margin debt. Rising margin debt fuels speculation; declining margin debt signals de-leveraging. Sustained declines of three or more months are historically associated with market corrections and bear markets.

Thresholds: Green if rising or flat; Yellow if declining 1–2 months; Red if declining 3+ consecutive months.

Update frequency: Monthly (FINRA data has approximately 6-week reporting lag).

Alert Levels

Level 0 — All Clear

No signals at Yellow or Red after persistence requirements are met. The system sees no evidence of bubble formation or unwind.

Level 1 — Early Warning

At least one Structural signal is Yellow or Red, or at least one Peak signal is Yellow. Early evidence of imbalance — monitor closely.

Level 2 — High Alert

Multiple signals across layers are elevated. The structural conditions for a bubble exist and market-level confirmation is beginning to appear.

Level 3 — Cascade Confirmed

All three layers are showing Red signals, or the Peak Override has triggered. The unwind is likely underway.

What This System Is NOT

It is important to be clear about what Bubble Trouble does not do:

Not a recession detector. Recessions can occur without tech bubbles, and tech bubbles can deflate without causing recessions.
Not a Fed policy tracker. While interest rates affect valuations, the system specifically distinguishes between macro-driven and sector-driven stress.
Not a general market timer. This system monitors one specific sector dynamic — AI infrastructure overinvestment — and nothing else.
Not predictive. The system is confirmatory: it identifies conditions that are present, not conditions that will occur in the future.
Not investment advice. This is an educational and analytical framework. See the full disclaimer.