The Business Problem
When NVIDIA surged 200%+ in 2023 and AI dominated every investment headline, a critical question emerged that billions of dollars hinged upon: Is this real value creation, or are we watching dot-com 2.0 unfold in real-time?
Traditional approaches failed to answer this:
- Valuation metrics (P/E ratios) break down during paradigm shifts
- Media narratives offer opinions, not data-driven answers
- No systematic framework existed to test: "Does this statistically resemble past bubbles?"
What I Built
An econometric detection system analyzing the "Magnificent Seven" tech stocks plus NASDAQ Composite Index to identify whether their valuations show statistical signatures of speculative bubbles.
| Ticker | Company | Sector Focus |
|---|
| AAPL | Apple Inc. | Consumer Electronics |
| AMZN | Amazon.com Inc. | E-commerce/Cloud |
| GOOG | Alphabet Inc. | Search/AI |
| META | Meta Platforms Inc. | Social/Metaverse |
| MSFT | Microsoft Corp. | Software/Cloud/AI |
| NVDA | NVIDIA Corp. | AI Infrastructure |
| TSLA | Tesla Inc. | EV/Energy |
| ^IXIC | NASDAQ Composite | Market Benchmark |
Methodology: Phillips-Shi-Yu Framework
I implemented the gold-standard econometric methodology for bubble detection—the same framework economists used to study the 2008 housing crisis and 1990s dot-com crash (Phillips, Shi & Yu, 2011, 2015).
5-Step Detection Process
1Collect Data: Download 15 years of monthly stock prices (2010-2025) from Yahoo Finance + Fama-French factor libraries
2Transform Prices: Convert raw prices to log prices, normalize to Base=100
3Run BSADF Test: For each point in time, run statistical test asking: "Is this price behavior explosive?"
4Compare to Threshold: Generate critical values using 150 wild bootstrap simulations with Rademacher multipliers
5Date-Stamp Bubbles: Mark when bubbles START and END, apply 6-month minimum duration filter
Statistical Engine Specifications
| Parameter | Value | Purpose |
|---|
| Total ADF Tests | 19,500+ per stock | Comprehensive coverage |
| Bootstrap Replications | 150 | Robust critical value estimation |
| Rolling Windows | ~130 per stock | Temporal analysis |
| Min Bubble Duration | 6 months | False positive filter |
| Confidence Level | 95% | Statistical significance |
| Data Period | Jan 2010 – Sep 2025 | 15+ years / 191 months |
Fama-French Factor Decomposition
Separated returns into components to understand what drives each stock:
- Alpha: Excess return beyond market factors — indicates company-specific value creation or speculative premium
- Beta: Market sensitivity — how much the stock amplifies market movements
- R²: Variance explained by market — distinguishes market-driven vs. company-specific behavior
Key Findings
17
Bubbles Detected
191
Months Analyzed
8
Securities Analyzed
95%
Confidence Level
8
Securities Covered
150
Bootstrap Replications
Complete Stock-by-Stock Results
| Stock | Alpha (Ann.) | Beta | R² | Bubbles | Risk Profile |
|---|
| TSLA | 18.9% | 1.29 | 0.24 | 2 | Highest alpha, lowest market correlation |
| NVDA | 14.7% | 1.77 | 0.39 | 2 | AI king, highest volatility |
| AAPL | 8.8% | 1.11 | 0.40 | 2 | Solid, moderate risk |
| META | 5.9% | 1.16 | 0.23 | 3 | Only stock with double AI-era bubbles |
| AMZN | 5.5% | 1.22 | 0.51 | 2 | Moderate across all metrics |
| MSFT | 4.7% | 1.06 | 0.53 | 2 | Most stable Mag-7 stock |
| GOOG | 4.2% | 1.09 | 0.41 | 2 | Balanced profile |
| ^IXIC | 0.1% | 1.09 | 0.95 | 2 | Benchmark (tracks market) |
Five Major Discoveries
- Universal Bubble Presence: All eight securities exhibit statistically significant bubble behavior in BOTH the pre-AI period (2010-2022) AND the AI boom era (2022-2025)
- NVIDIA Leads in Explosive Growth: With 40,000%+ returns since 2010, 14.7% annualized alpha, and the highest market beta (1.77), NVIDIA shows the most aggressive growth characteristics
- Tesla Generates Highest Alpha: Despite market volatility, Tesla produces 18.9% annualized excess returns—the highest among all analyzed securities—yet shows one of the lowest R² values (0.24)
- META Shows Multiple AI-Era Bubbles: Uniquely among the Magnificent Seven, META exhibits two distinct bubble periods during the AI boom (Dec 2022–Sep 2025), suggesting heightened speculative activity around its AI/metaverse pivot
- Market-Wide Phenomenon: The NASDAQ index mirrors individual stock bubble patterns, indicating this is systematic rather than idiosyncratic speculative behavior—the entire tech sector is moving together
Understanding the Metrics (Plain English)
Alpha — "The Extra Juice"
What it measures: Returns beyond what the overall market gives you.
Think of it like: You invested in a stock and the market went up 10%. If your stock went up 15%, that extra 5% is roughly your alpha—the "bonus" you got.
Why Tesla's 18.9% alpha matters: Tesla returned almost 19% extra per year on top of market returns. That's either incredible value creation... or a speculative premium waiting to correct.
Beta — "The Roller Coaster Factor"
What it measures: How much the stock swings compared to the market.
- Beta = 1.0 → Moves exactly with market
- Beta > 1.0 → Amplifies market movements
- Beta < 1.0 → Dampens market movements
Why NVIDIA's 1.77 beta matters: When the market goes up 10%, NVIDIA goes up 17.7%. But when the market drops 10%, NVIDIA drops 17.7%. High beta = high risk, high reward.
R² — "The Independence Score"
What it measures: What percentage of the stock's movement is explained by the overall market.
- NASDAQ's 0.95 R²: 95% of its movement is just the market moving
- META's 0.23 R²: Only 23% is market-driven—the other 77% is company-specific factors (or speculation)
Low R² + High Alpha = Either genuine innovation OR speculative bubble behavior
Technical Stack
Python 3.x
pandas
NumPy
statsmodels
scipy
scikit-learn
yfinance
matplotlib
seaborn
Streamlit
Data Sources
- Price Data: Yahoo Finance (monthly OHLC)
- Factor Data: Kenneth French Data Library (Fama-French factors)
Deliverables
- Interactive Streamlit Dashboard
- 8 publication-quality PNG visualizations
- CSV summaries of bubble periods
- Fama-French regression results
- HBS-style PDF research report
Investment Implications
The evidence suggests that current AI stock valuations contain a significant speculative component that extends beyond fundamental value creation.
Historical precedent: Bubble periods tend to end with substantial corrections
Risk differentiation: High R² stocks (MSFT, AMZN) are more market-driven and "safer"; Low R² stocks (TSLA, META) carry more idiosyncratic risk
This analysis does not constitute investment advice but provides a quantitative framework for risk assessment.
Skills Demonstrated
| Category | Skills |
|---|
| Technical | Econometric modeling, time series analysis, statistical hypothesis testing, bootstrap inference, API integration, data visualization |
| Business | Quantitative research design, financial analysis, technical writing, complex concept simplification, risk assessment |
| Tools | Python, pandas, NumPy, statsmodels, scipy, matplotlib, seaborn, Streamlit, Git |
What I Learned
Building this system taught me that the best business decisions sit at the intersection of quantitative rigor and strategic thinking. Markets tell stories through data. My job was to listen.
The difference between "transformative technology" and "speculative mania" isn't philosophical—it's measurable. This project gave me the tools to measure it.
Evidence Chain
L1
Live Dashboard
Interactive Streamlit visualization with all charts
Available
L2
Methodology + Raw Data
psy_bubble_summary.csv, fama_french_loadings_psy.csv
Available
L3
Research Walkthrough
Full methodology call + HBS-style PDF report
On Request
