Methodology
How we track AI bubble risk based on historical patterns from tech booms and busts.
📚 Inspiration
Our methodology is inspired by (but not identical to) Robinhood's analysis of bubble indicators, which in turn references the historical patterns documented in Devil Take the Hindmost: A History of Financial Speculation by Edward Chancellor.
We've adapted these concepts specifically for tracking AI sector risk using quantitative metrics and semi-automated data collection.
🎯 Key Principle: Score-Based Ratings
All traffic light ratings (🟢 green, 🟡 yellow, 🔴 red) are determined purely from numerical scores—not from separate domain-specific rules.
How it works:
- Calculate a score (0-100) for each metric based on underlying financial data
- Compare the score to predefined thresholds (YELLOW, RED)
- Assign the rating based solely on the score
This ensures consistency: the same score always produces the same rating, and the thresholds are calibrated to historical bubble patterns.
Overall Bubble Risk Index (0–100)
We combine three metrics into a single risk score using a weighted average:
- Valuations: 40% weight
- Financing Mix: 35% weight
- Hot IPOs: 25% weight
Calculation Formula
Overall Score =
Valuations Score × 0.40 +
Financing Score × 0.35 +
IPOs Score × 0.25
Overall Thresholds (Dynamically Calculated)
The overall thresholds are not arbitrary—they are derived from the individual metric thresholds using the same weighted formula:
Yellow Threshold =
33 × 0.40 + 37 × 0.35 + 30 × 0.25 = 34
Red Threshold =
65 × 0.40 + 63 × 0.35 + 55 × 0.25 = 62
Traffic light zones:
- 🟢 Green: < 34 (Low risk)
- 🟡 Yellow: 34–61 (Watch zone)
- 🔴 Red: ≥ 62 (Elevated risk)
1. Valuations
Bubbles form when prices surge beyond plausible future cash flows. We track forward P/E and PEG ratios for a basket of AI bellwether stocks.
Universe
NVDA, MSFT, GOOGL, META, AVGO, AMD, ARM, SMCI (configurable)
What we compute
- Forward P/E (price / next-12-month earnings estimate)
- PEG (forward P/E ÷ EPS growth rate %)
- Basket median for each metric
Scoring Formula
P/E Score = ((Forward P/E - 15) / (60 - 15)) × 100
PEG Score = ((PEG - 0.5) / (4.0 - 0.5)) × 100
Final Score = (P/E Score + PEG Score) / 2
Range: 0-100 (capped at min/max)
Traffic Light Ratings (Score-Based)
The traffic light rating is determined from the score using these thresholds:
- 🟢 Green: Score < 33
- 🟡 Yellow: Score 33–64
- 🔴 Red: Score ≥ 65
Example: A P/E of 30 and PEG of 2.0 produces a score around 38, triggering yellow. A P/E of 50 and PEG of 3.0 produces a score around 75, triggering red. Thresholds are calibrated based on historical bubble patterns (dot-com, 2021 tech bubble).
2. Financing the Boom
When debt heavily funds capex, fragility increases. We monitor hyperscaler spending patterns, leverage, and debt issuance.
Universe
MSFT, AMZN (AWS), GOOGL, META (the major cloud/AI infrastructure spenders)
What we compute
- Capex/FCF Ratio (TTM capex ÷ free cash flow)
- Net Debt / EBITDA (leverage indicator)
- Debt Issuance Momentum (YTD or TTM change in total debt)
Scoring Formula
Capex Score = (Capex/FCF / 2.5) × 100 × 0.5
Leverage Score = (max(0, Net Debt/EBITDA) / 3.0) × 100 × 0.3
Issuance Score = ((Debt Change % + 10) / 60) × 100 × 0.2
Final Score = Capex Score + Leverage Score + Issuance Score
Weights: Capex/FCF 50%, Leverage 30%, Issuance 20%
Traffic Light Ratings (Score-Based)
The traffic light rating is determined from the score using these thresholds:
- 🟢 Green: Score < 37
- 🟡 Yellow: Score 37–62
- 🔴 Red: Score ≥ 63
Example: Capex/FCF of 1.0, Debt/EBITDA of 1.0, and 10% debt growth produces a score around 37, triggering yellow. Capex/FCF of 1.5, Debt/EBITDA of 2.0, and 30% debt growth produces a score around 63, triggering red.
3. Hot IPOs
Extreme IPO volume and first-day pops are classic late-cycle tells. We track trailing-90-day tech IPO activity.
What we compute
- Trailing-90D tech IPO count (US)
- Trailing-90D proceeds
- Average first-day return
Scoring Formula
Count Score = (90D Count / (Post-2000 Median × 2)) × 100
Return Score = (Avg First-Day % / 50) × 100
Final Score = (Count Score + Return Score) / 2
Post-2000 Median ≈ 50 IPOs per 90 days
Traffic Light Ratings (Score-Based)
The traffic light rating is determined from the score using these thresholds:
- 🟢 Green: Score < 30
- 🟡 Yellow: Score 30–54
- 🔴 Red: Score ≥ 55
Example: 25 IPOs (50% of median) with 15% first-day returns produces a score around 30, triggering yellow. 50 IPOs (100% of median) with 30% first-day returns produces a score around 55, triggering red. The 30% first-day pop threshold mirrors late-1990s exuberance.
Alert Triggers
We send email alerts when our signals indicate elevated burst risk:
- Overall Red for 2 consecutive runs, OR
- Valuations Red AND Financing Red simultaneously, OR
- IPOs Red for 2 consecutive runs with Valuations or Financing at Red/Yellow
Cool-down: Minimum 14 days between alert emails.
Data Sources
- Valuations: Public financial data providers (Yahoo Finance, company IR)
- Financing: Public financial data providers (Yahoo Finance, company IR)
- IPOs: Renaissance Capital, NASDAQ IPO Center, Jay Ritter database