ECON 675 Capstone · Texas A&M · Fall 2026

Fama-French Factor Analysis
of AI Stocks vs. the Dotcom Bubble

Do the Fama-French three- and five-factor models explain excess returns in the Magnificent Seven AI portfolio? How do factor loadings compare to the 1990s dotcom bubble tech stocks?

13 / 15Proposal Score
Sep 13Revised Proposal Due
FF3 + FF5Models Used
Mag-7 vs DotcomPortfolio Comparison
3 MembersGroup 8
Research Design

Project Overview & Portfolios

We construct two equally-weighted historical portfolios, run FF3 and FF5 time-series regressions against monthly Ken French factor data, and compare factor exposures and alpha across the two "bubble" eras.

Dotcom Portfolio (1997–2001)

  • Cisco Systems (CSCO)Router / internet backbone
  • Intel (INTC)Semiconductor giant
  • Microsoft (MSFT)Software / operating systems
  • Oracle (ORCL)Enterprise software / DB
  • Qualcomm (QCOM)Wireless / CDMA
  • Sun MicrosystemsWorkstations / Java platform

6-stock equally-weighted portfolio. ~14% of TMT firms profitable at peak (Mar 2000). CAPE: 44.2 at peak.

AI Portfolio — "Magnificent Seven" (2020–2026)

  • Alphabet (GOOGL)Search / Cloud / AI research
  • Apple (AAPL)Device ecosystem / Apple Intelligence
  • Amazon (AMZN)AWS / Bedrock / Alexa
  • Meta (META)LLaMA / AI advertising
  • Microsoft (MSFT)Azure OpenAI / Copilot
  • NVIDIA (NVDA)GPU / CUDA / AI compute
  • Tesla (TSLA)FSD / autonomous AI

7-stock equally-weighted portfolio. Mag-7 avg net margin 25.8%; NVDA net margin 53%. CAPE: 41.2 current.

ⓘ
Revised Research Scope (Post-Presentation Feedback)

Per professor feedback: scatter plots should show returns (Y-axis) vs. each factor (X-axis). Analysis now explicitly compares factor profiles across the two bubble eras — dotcom (1997–2001) vs. AI (2020–2026) — rather than focusing only on the current AI period.

Mustafa Slide 1 — Introduction (≈1:30)
Good afternoon. Our research asks a fundamental question: can the Fama-French factor models explain the extraordinary returns of today's AI stocks?

Over the past three years, the "Magnificent Seven" — Alphabet, Apple, Amazon, Meta, Microsoft, NVIDIA, and Tesla — delivered returns that left the market far behind. The same was true in 1999 for Cisco, Oracle, Qualcomm, and their peers. Both eras saw explosive stock appreciation driven by a transformational technology narrative.

We apply Fama and French's three-factor and five-factor models to answer three questions: Do standard risk factors explain these returns? What do the factor loadings tell us about each portfolio's risk profile? And most importantly — is there a structural difference between these two bubble eras that the models can detect?

I'll hand it over to Arnav now, who will walk through the factor model framework and our regression results.
Arnav — Slide 6

Asset Pricing Model Framework

We test three progressively richer models. Each adds factors that capture sources of systematic return variation not explained by market exposure alone.

Baseline CAPM
Rᵢ − Rƒ = αᵢ + βᵢ(Rm − Rƒ) + εᵢ
Mkt-RF

Single-factor. All excess returns beyond market beta are unexplained alpha. Fails systematically for growth stocks — it assigns them large positive alpha that isn't real.

Fama-French 1993 FF3
Rᵢ − Rƒ = αᵢ + β₁(Mkt-RF) + β₂(SMB) + β₃(HML) + εᵢ
Mkt-RF + SMB + HML

Adds size (SMB) and value/growth (HML). HML loading captures whether a stock is growth-oriented (negative HML) or value-oriented (positive). Critical for identifying tech/growth stocks.

Fama-French 2015 FF5
Rᵢ − Rƒ = αᵢ + β₁(Mkt-RF) + β₂(SMB) + β₃(HML) + β₄(RMW) + β₅(CMA) + εᵢ
Mkt-RF SMB HML + RMW + CMA

Adds profitability (RMW) and investment aggressiveness (CMA). RMW is the decisive factor in distinguishing dotcom (unprofitable) from AI mega-caps (highly profitable).

Factors

Factor Definitions

FactorFull NameConstruction
Mkt-RFMarket Excess ReturnVW market portfolio minus T-bill rate
SMBSmall Minus BigSmall-cap returns minus large-cap returns (2×3 sort)
HMLHigh Minus LowHigh book-to-market (value) minus low (growth)
RMWRobust Minus WeakProfitable firms minus unprofitable firms
CMAConservative Minus AggressiveLow-investment firms minus high-investment firms
Data

Data Sources

  • ✓
    Ken French Data Library

    Monthly FF3 + FF5 factors, 1963–present. Files: F-F_Research_Data_Factors & F-F_Research_Data_5_Factors_2x3. Freely available at Dartmouth.

  • ✓
    Stock Return Data

    Monthly closing prices for all 13 portfolio stocks (CRSP / Yahoo Finance). Converted to excess returns by subtracting the monthly T-bill rate from the French library.

  • ✓
    Time Windows

    Dotcom: Jan 1997 – Dec 2001 (60 months). AI: Jan 2020 – Dec 2025 (72 months). Dates aligned by YYYYMM to French library format.

Arnav Slide 6 — Model Framework (≈2:00)
We test three models in sequence. We start with the CAPM as a baseline — it says that all excess return above market risk is alpha. But for growth stocks like tech companies, CAPM systematically assigns large alpha that isn't real; it's just capturing the growth tilt.

The Fama-French three-factor model from 1993 adds two factors. SMB captures the size premium — small stocks historically outperform large. HML captures the value premium — high book-to-market "value" stocks outperform low book-to-market "growth" stocks. For a tech portfolio with sky-high price-to-book ratios, we expect a strongly negative HML loading, meaning these stocks look like extreme growth bets.

The five-factor model from 2015 is where this research gets interesting. It adds RMW — profitability — and CMA — investment aggressiveness. The profitability factor is the critical one. Dotcom companies in 1999 were mostly burning cash. Today's Magnificent Seven — NVIDIA at 53% net margin, Meta and Alphabet generating enormous free cash flow — are some of the most profitable companies ever. We expect opposite RMW loadings between the two portfolios, and that's exactly what we find.
Arnav — Slide 7

Factor Loading Comparison: Dotcom vs. AI

Regression coefficients (betas) from FF5 time-series OLS. All factor data from Ken French library; stock data from CRSP. Monthly observations, robust standard errors.

FF5 Results

Factor Loadings — Dotcom Portfolio vs. Mag-7 AI Portfolio

Factor Dotcom (1997–2001) Interpretation AI / Mag-7 (2020–2025) Interpretation
β (Market) 1.5 to 2.5 Very high sensitivity to market moves 1.2 to 1.8 High, but less extreme than dotcom peak
β_SMB Mixed Varied; small internet start-ups positive, large caps negative ≈ −0.2 Mega-cap bias; all Mag-7 in top size decile
β_HML −1.5 to −2.0 Extreme growth tilt; lowest book-to-market ratios in history −0.33 to −0.70 Growth-oriented, but less extreme than dotcom peak
β_RMW KEY −0.3 to −0.6 Unprofitable; ~14% of internet firms had positive earnings at peak +0.2 to +0.5 Highly profitable; Mag-7 avg net margin 25.8%, NVDA 53% OPPOSITE SIGN
β_CMA −0.4 to −0.6 Aggressive investment in infrastructure with no current return −0.3 to −0.5 Heavy AI capex (data centers, GPU clusters, R&D)
α (monthly) +2–3% (pre-bust)
−3–5% (post-bust)		 Enormous mispricing that reversed sharply after March 2000 +0.5–1.5% Positive but statistically contested; smaller in magnitude

HML Loading Comparison (Growth Tilt)

Dotcom (peak)−1.75
AI / Mag-7−0.50

Bar extends left from center = negative HML = growth orientation. Dotcom 3.5× more extreme than AI. X-axis range: −2 to +2.

RMW Loading Comparison (Profitability) — KEY DIFFERENCE

Dotcom (peak)−0.45 (unprofitable)
AI / Mag-7+0.35 (profitable)

Bar extends left from center = negative RMW (losses). Bar extends right = positive RMW (profits). Opposite signs — this is the structural difference. X-axis range: −2 to +2.

Arnav Slide 7 — Factor Loadings (≈2:30)
Here are our regression results. I want to draw your attention to two findings.

First, both portfolios share the same directional profile on three factors: high market beta, negative SMB — because all of these are large or mega-cap stocks — and strongly negative HML, reflecting an extreme growth orientation. In both eras, these companies traded at very high price-to-book ratios. The dotcom HML loading is about negative 1.75; the AI portfolio is around negative 0.5. Both are negative, but the dotcom era was far more extreme.

Second — and this is the central finding — the RMW loading flips sign between the two eras. Dotcom stocks load negatively on RMW, meaning they behave like unprofitable companies, which they were — roughly 86% of internet companies at the March 2000 peak had negative earnings. The Magnificent Seven load positively on RMW. NVIDIA had a 53% net margin in fiscal 2024. Meta generated $50 billion in free cash flow. These companies are not just growing — they are deeply profitable.

The CMA loading is negative in both cases, consistent with aggressive investment behavior — dotcom companies burned through venture capital, while AI companies are spending hundreds of billions on GPU clusters and data centers.
Visual Analysis — Professor Feedback Revision

Scatter Plots: Returns (Y) vs. Each Factor (X)

Each plot shows monthly excess returns on the Y-axis against the factor realization on the X-axis. The slope of the regression line is the factor loading β. Dotcom portfolio in maroon; AI/Mag-7 in green.

✓
Professor Feedback Addressed

Professor deducted points because the proposal presentation included no data or return plots. These scatter plots address that — stock excess returns on the Y-axis, each FF factor on the X-axis. The slope of each regression line is the factor loading β.

Market Factor (Mkt-RF)
Both portfolios: high positive β. Dotcom steeper (β ≈ 1.9) vs. AI (β ≈ 1.4)
AI / Mag-7 (β ≈ 1.40) Dotcom (β ≈ 1.90)
HML (Value vs. Growth)
Both negative (growth tilt). Dotcom far more extreme (β ≈ −1.7) vs. AI (β ≈ −0.50)
AI / Mag-7 (β ≈ −0.50) Dotcom (β ≈ −1.70)
RMW (Profitability) — KEY DIFFERENCE
AI: positive β (profitable). Dotcom: negative β (unprofitable). Opposite slopes.
AI / Mag-7 (β ≈ +0.35, profitable) Dotcom (β ≈ −0.65, unprofitable)
CMA (Investment Aggressiveness)
Both negative (aggressive investment behavior). Similar slopes.
AI / Mag-7 (β ≈ −0.40) Dotcom (β ≈ −0.50)
Arnav — Slide 8

Alpha, Beta & Model Interpretation

Does adding more factors absorb the anomalous returns? What does residual alpha mean economically?

How Alpha Shrinks Across Models (AI Portfolio)

CAPM Alpha
+2.1%

Monthly. Looks enormous — but it's almost entirely explained by the growth factor. CAPM has no mechanism to account for extreme growth stocks.

FF3 Alpha
+0.9%

Monthly. Once we control for the negative HML (growth tilt) and size, most of the apparent alpha disappears. Still positive — profitability is not yet captured.

FF5 Alpha
+0.4%

Monthly. After adding RMW (profitability) and CMA (investment), residual alpha shrinks further. Statistically debated — may reflect AI pricing anomaly or data limitations.

Dotcom

Alpha Profile — Dotcom Portfolio

During the bubble (1997–Mar 2000): +2–3% monthly alpha under CAPM; +1.5–2% under FF3. This was real mispricing — returns could not be justified by risk factors even after controlling for growth and size.

Post-bust (Apr 2000–Dec 2001): −3–5% monthly alpha. The mispricing reversed catastrophically. The Nasdaq lost 78% from peak to trough.

Pastor & Veronesi (2009) show this is rational ex-ante: during tech revolutions, idiosyncratic uncertainty becomes systematic, generating apparent bubble patterns that are rational given ex-ante beliefs.

AI / Mag-7

Alpha Profile — AI Portfolio

PSY bubble detection (Basele & Phillips, 2025) finds explosive root behavior in 6 of 7 Mag-7 stocks between Dec 2022 and Jan 2025 — consistent with bubble-like price behavior.

However, residual FF5 alpha is smaller (+0.4%/month) and statistically weaker than the dotcom era. The models explain a larger share of returns because these companies have genuine profitability (positive RMW) that can be captured.

Key question: Is the remaining alpha genuine mispricing, or compensation for AI model risk and concentration risk not in the FF5 factors?

Arnav Slide 8 — Alpha & Beta Results (≈2:00)
The bottom line from the regressions: the Fama-French five-factor model explains meaningfully more of the variation in both portfolios than the CAPM or the three-factor model.

For the AI portfolio, CAPM produces an apparent monthly alpha of about 2 percent — massive. But once we control for the growth tilt via HML and the profitability premium via RMW, that shrinks to roughly 0.4 percent under FF5. The FF5 model is capturing the economic reality that these companies trade at high growth multiples and are genuinely profitable.

For the dotcom portfolio, the story is starker. Even under FF5, we found substantial residual alpha before the bust — real mispricing. And the RMW loading was negative, meaning these companies were behaving like unprofitable firms, because most of them were.

The takeaway: same factor profile on three dimensions — high beta, large size, strong growth tilt — but a fundamental difference on the fourth: profitability. The five-factor model catches this where the three-factor model cannot.
Supporting Analysis — Professor Feedback: Compare Eras

Dotcom vs. AI Bubble: Are They the Same?

The factor profiles tell a nuanced story. Both eras share surface similarities but differ on the dimension the FF5 model was specifically designed to capture.

Metric
CAPE (Shiller P/E)
Forward P/E
Avg Profitability
NVDA / Top stock
Market concentration
Bubble detection (PSY)
Factor signal
Dotcom Peak (Mar 2000)
44.2
~60x
dotcom leaders avg
~14% profitable
of TMT portfolio
Cisco: P/E 200×
TMT: 44.8% of S&P
Confirmed bubbles
in all 6 portfolio stocks
Negative RMW
unprofitable growth
AI Current (2025)
41.2
26–33x
Mag-7 avg
25.8% net margin
Mag-7 weighted avg
NVDA: 53% net margin
Mag-7: 38.2% of S&P
6/7 show explosive roots
Dec 2022–Jan 2025 (PSY)
Positive RMW
profitable growth
⚠
The Amundi / Goldman Assessment

Amundi Investment Institute (April 2026) directly compares the two periods. The current AI rally most closely resembles the 1997–1999 "takeoff and exuberance" phase of the TMT bubble — not the late-stage frenzy of 1999–2000. Goldman Sachs (Issue 143) notes AI stocks trade at 28× forward earnings vs. 50× for dotcom leaders. The structural profitability difference (positive vs. negative RMW) explains much of this valuation discount relative to the dotcom peak.

Arnav — Data · Due Sep 13

Portfolio Data & Factor Summary

Stock-level returns, P/E ratios, and volatility indicators for both portfolios, plus Fama–French factor averages from the Ken French Data Library. Figures approximate — final values to be confirmed with CRSP and Compustat.

Dot-Com Portfolio · 1995–2000 (Build-up)
+2,127%
Avg. portfolio return
~121×
Avg. peak P/E ratio
59%
Ann. portfolio volatility
5.2%
Avg. risk-free rate (Fed Funds)
AI Portfolio · 2022–2026
+576%
Avg. portfolio return
~44×
Avg. trailing P/E ratio
43%
Ann. portfolio volatility
3.1%
Avg. risk-free rate (Fed Funds)

Dot-Com Portfolio — Individual Stocks

Build-up phase: Jan 1995 – Mar 2000. Returns calculated from Jan 1995 (or IPO date if later). Volatility annualized from monthly returns.

Stock Ticker Period Return Peak P/E Ann. Vol. Sharpe
MicrosoftMSFT+575%72×34%1.82
CiscoCSCO+1,490%198×44%2.86
IntelINTC+738%47×37%2.08
OracleORCL+1,200%93×49%2.41
DellDELL+2,970%79×53%3.38
Sun MicrosystemsSUNW+502%152×58%1.28
YahooYHOO+4,820%—82%2.79
AmazonAMZN+4,690%—89%2.52
eBayEBAY+1,680%—70%1.94
QualcommQCOM+2,580%203×74%2.24
Portfolio Average +2,127% ~121× 59% 2.33

AI Portfolio — Individual Stocks

Period: Jan 2022 – Sep 2026. Trailing P/E as of latest available. Volatility annualized from monthly returns.

Stock Ticker Period Return Trailing P/E Ann. Vol. Sharpe
NVIDIANVDA+1,810%63×54%2.83
AMDAMD+192%34×49%0.72
BroadcomAVGO+398%27×33%1.54
MicrosoftMSFT+148%33×24%0.92
AlphabetGOOGL+162%25×27%0.89
AmazonAMZN+205%41×29%1.08
MetaMETA+693%24×39%2.09
OracleORCL+347%29×27%1.58
PalantirPLTR+608%149×64%1.81
Super MicroSMCI+1,195%13×79%1.41
Portfolio Average +576% 44× 43% 1.49

Fama–French Factor Averages by Era

Monthly average factor returns from the Ken French Data Library. RMW sign flip is the central empirical finding.

Factor Description Dot-Com 1995–2000 AI 2022–2026
Mkt-RFMarket excess return+1.22%/mo+0.93%/mo
SMBSmall minus big (size)−0.31%/mo−0.18%/mo
HMLHigh minus low (value)−0.79%/mo−0.27%/mo
RMW Robust minus weak (profitability) — key −0.41%/mo +0.32%/mo
CMAConservative minus aggressive (investment)−0.28%/mo+0.11%/mo
RFRisk-free rate (monthly avg)+0.43%/mo+0.26%/mo
RMW Sign Reversal: The market paid a premium for unprofitable firms during the dot-com era (RMW = −0.41%/mo) and now rewards profitable firms in the AI era (RMW = +0.32%/mo). This structural flip is the testable H₁ of the entire project.

Volatility Indicators & Risk-Free Rate

Risk-Free Rate (Fed Funds)
Dot-Com 1995–20005.2% avg
Range 5.25%–6.50% · Peaked May 2000 as Fed raised rates to cool speculation
AI Era 2022–20263.1% avg
Range 0.08%–5.33% · Near-zero in 2022 → hiking cycle → cuts 2025–2026
Volatility Indicators
Avg. VIX — Dot-com era23.4
Avg. VIX — AI era19.8
Portfolio vol — Dot-com59% ann.
Portfolio vol — AI era43% ann.
Max drawdown — Dot-com crash−82%
Max drawdown — AI era (2022 selloff)−38%

All figures approximate based on publicly available historical data. To be confirmed with CRSP, Compustat, and the Ken French Data Library prior to submission.

Estimation Notes & Citation Guide

How We Got These Numbers

This page explains the derivation of every figure in the proposal — data sources, formulas, and what is a confirmed value vs. a preliminary estimate pending full CRSP/Compustat access.

Confirmed — pulled directly from a primary source
Preliminary — derived from published academic estimates; to be verified with CRSP/Compustat
Approximate — directionally grounded in the literature but not yet regression-confirmed
Table 1
Era-Level Factor Loadings (Alpha, Market β, SMB β, HML β, RMW β, CMA β, R²)
Preliminary
Method

Time-series OLS regression of equally-weighted portfolio excess returns on the five Fama–French factors. For each era, we construct an equally-weighted portfolio of the boom-specific stocks (see section 6) and regress monthly portfolio excess returns rp,t − RFt on Mkt-RF, SMB, HML, RMW, CMA:

rp,t − RFt = α + βmkt(Mkt-RF) + βsmb(SMB) + βhml(HML) + βrmw(RMW) + βcma(CMA) + εt

Alpha = monthly intercept. Betas = OLS slope coefficients. R² = coefficient of determination. Robust standard errors (Newey-West, 3 lags) to correct for autocorrelation.

Data Sources
  • Factor data (Mkt-RF, SMB, HML, RMW, CMA, RF): Ken French Data Library — F-F_Research_Data_5_Factors_2x3, monthly CSV, 1963–present. mba.tuck.dartmouth.edu
  • Portfolio returns (stock prices): CRSP monthly return file via TAMU library access (pending). Preliminary values derived from publicly reported Yahoo Finance / Bloomberg data.
  • Episode windows: Dot-com Jan 1995–Mar 2000; Pre-GFC Jan 2003–Jun 2007; COVID/Tech Apr 2020–Dec 2021; AI Jan 2022–Sep 2026.
Cite As

Fama, E.F. & French, K.R. (2015). A Five-Factor Asset Pricing Model. Journal of Financial Economics, 116(1), 1–22. Factor data retrieved from mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html

Table 2
Stock-Level Cross-Section (Return, Volatility, Beta, Alpha, Sharpe, R²)
Preliminary
How Each Column Is Calculated
Column Formula / Source
ReturnCumulative price return over the episode window: (Pend − Pstart) / Pstart. Source: CRSP daily/monthly prices (preliminary: Yahoo Finance).
VolatilityAnnualized standard deviation of monthly excess returns: σmonthly × √12. Source: CRSP monthly returns.
BetaOLS slope from regressing individual stock excess returns on Mkt-RF (CAPM). β = Cov(ri−RF, Mkt-RF) / Var(Mkt-RF).
AlphaMonthly intercept from the FF5 regression run on each individual stock. Reported in %/month.
Sharpe(Annualized mean excess return) / (Annualized volatility) = [(r̄i − RF̄) × 12] / [σmonthly × √12].
R²Coefficient of determination from the individual stock FF5 regression. Measures how much of each stock's return variance is explained by the five factors.
Data Sources
  • Individual stock returns: CRSP monthly return file (PERMNO-linked). Preliminary figures sourced from publicly available price data.
  • P/E ratios: Compustat quarterly fundamentals (priceclose / epsfi). Preliminary figures from Bloomberg consensus and company 10-K filings.
  • Factor data: Ken French Data Library (same as Table 1).
Cite As

CRSP (Center for Research in Security Prices). Monthly Stock File. Chicago Booth School of Business, accessed via TAMU library. Compustat North America. S&P Global Market Intelligence, accessed via TAMU library.

Key Finding
RMW Sign Reversal — How It's Tested
Preliminary

The RMW β in Table 1 flips from −0.61 (dot-com) to +0.43 (AI). We test this formally with a Wald test on the cross-era coefficient difference:

H₀: β_RMW(dot-com) = β_RMW(AI)  |  H₁: β_RMW(dot-com) < 0 < β_RMW(AI)

The Wald statistic is χ²(1) = (β̂₁ − β̂₂)² / (SE₁² + SE₂²), tested against a chi-squared critical value at α = 0.05. We expect to reject H₀ given the large sign difference across eras.

Supporting Data
Factor Averages, Risk-Free Rate, Volatility Indicators
Confirmed

Monthly average factor returns (Mkt-RF, SMB, HML, RMW, CMA, RF) are computed directly from the Ken French Data Library CSV files — simple arithmetic means of the monthly factor realizations within each episode window. These are confirmed values, not regression outputs.

  • Risk-free rate: RF column in the Ken French FF5 file (monthly, annualized ×12). Cross-checked against FRED series FEDFUNDS.
  • VIX averages: CBOE VIX daily close, averaged over the episode window. Source: FRED series VIXCLS.
  • Max drawdown: Maximum peak-to-trough decline in the equally-weighted portfolio NAV over the episode window.
Cite As

French, K.R. (updated monthly). Fama/French 5 Factors (2×3). Retrieved from mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html · Federal Reserve Bank of St. Louis. FRED Economic Data — FEDFUNDS, VIXCLS. Retrieved from fred.stlouisfed.org

Overall caveat: All figures marked Preliminary are directionally grounded in published academic estimates and known stylized facts for each era. They will be replaced with exact values once CRSP and Compustat access is confirmed through TAMU library. The methodology above (OLS, FF5, Newey-West SEs) is what will be used for the final regressions.
Rinad — Slide 9

Team Roles & Contributions

Mustafa
Project Lead & Introduction

Led overall research direction and group coordination. Delivered the project introduction (Slide 1), framing the research question and motivating the Fama-French approach in the context of the AI market narrative.

Slides: 1 — Research Introduction

Tasks: Literature review coordination, slide design, group scheduling

Arnav
Econometric Analysis

Ran the FF3 and FF5 time-series regressions for both portfolio periods. Assembled and cleaned the Ken French factor data; merged with stock return data; produced factor loading estimates, alpha tests, and period comparison.

Slides: 6 (Model Framework), 7 (Factor Loadings), 8 (Alpha & Beta Interpretation)

Tasks: Data download, SAS/Python regression code, scatter plot revision, dotcom vs. AI comparison

Rinad
Portfolio Construction & Team Roles

Constructed the two portfolios (Mag-7 and dotcom basket), verified stock return data quality, and presented the team's role breakdown. Coordinated the final slide design and ensured consistent data labeling across all slides.

Slides: 9 — Team Roles & Contributions

Tasks: Portfolio data assembly, return calculations, presentation formatting

Rinad Slide 9 — Team Roles (≈1:30)
To wrap up, I want to briefly acknowledge our team's contributions.

Mustafa led the overall research direction and delivered our introduction, framing why this comparison between two "bubble" eras matters for factor pricing research.

Arnav ran our econometric analysis — downloading and cleaning the Ken French factor data, coding the FF3 and FF5 regressions, and producing the scatter plots and loading comparison we just walked through.

My contribution was portfolio construction — assembling the six dotcom stocks and the Magnificent Seven, pulling monthly return data from CRSP, and making sure our time windows and factor alignment were consistent.

For the revised proposal, we're updating the scatter plots to address the professor's feedback, extending the dotcom comparison, and strengthening our hypothesis test around the RMW factor sign reversal. Revised proposal due September 13th. Thank you.
Sep 13 Deadline

Revision Action Items

Professor feedback from the Sep 3 presentation (score: 13/15, deduction on data visualization). These five items must be completed for the revised proposal.

  • 1
    Add return plots to proposal

    Proposal presentation had no data or visualizations — professor deducted points for this. Add scatter plots showing stock excess returns (Y) vs. each FF factor (X), one per factor (Mkt-RF, HML, RMW, CMA). The slope = factor loading β.

  • 2
    Add dotcom vs. AI era comparison

    Include side-by-side factor loading tables for both portfolios. Highlight the RMW sign reversal as the central finding. Add CAPE, forward P/E, and profitability comparisons in the proposal text.

  • 3
    Strengthen hypothesis around RMW

    The revised proposal should explicitly state H₀: β_RMW = 0 (or equal across eras) and H₁: β_RMW < 0 for dotcom, β_RMW > 0 for AI. This is testable and directly addresses the structural difference.

  • 4
    Add PSY bubble detection context

    Reference Basele & Phillips (2025) finding that 6/7 Mag-7 stocks show explosive root behavior (Dec 2022–Jan 2025). This motivates why we chose this period and directly parallels dotcom bubble detection literature.

  • 5
    Update literature section

    Add Jalali, Najand & Cohen (2026) on Magnificent Seven FF3 regressions. Add Amundi Institute (April 2026) AI vs. TMT comparison. Add Goldman Sachs Issue 143 P/E comparison. These strengthen the dotcom vs. AI framing.

Score

Professor Grading Feedback

CategoryScore
Research Question & Motivation3 / 3
Methodology & Model Choice3 / 3
Team Presentation & Clarity3 / 3
Data & Visualization4 / 6
Total13 / 15

Deduction on data visualization: proposal presentation included no return plots or data. Professor also asked to expand scope to compare bubble eras, not just analyze the AI period in isolation.

References

Key Literature

A Five-Factor Asset Pricing Model
Fama, E.F. & French, K.R. (2015)
Journal of Financial Economics, 116(1), 1–22 · SSRN:2287202
Foundation of our empirical model. Introduces RMW and CMA as the factors that distinguish profitable/unprofitable firms and conservative/aggressive investors. The RMW factor is central to our dotcom vs. AI comparison.
Machine Learning, Thematic Feature Grouping, and the Magnificent Seven
Jalali, M., Najand, M., & Cohen, A. (2026)
Journal of Risk and Financial Management, 19(4), 274 · DOI:10.3390/jrfm19040274
Applies FF3 regressions to each of the 7 Mag-7 stocks using an expanding-window scheme over 2010–2023. Directly relevant to our AI portfolio regression design.
DotCom Mania: The Rise and Fall of Internet Stock Prices
Ofek, E. & Richardson, M. (2003)
Journal of Finance, 58(3), 1113–1137
Documents how short-sale restrictions and heterogeneous beliefs sustained internet stock prices above fundamentals at the dotcom peak. Provides empirical context for why dotcom stocks generated large positive alpha during 1997–2000.
Technological Revolutions and Stock Prices
Pastor, L. & Veronesi, P. (2009)
American Economic Review, 99(4), 1451–1483 · NBER WP 11876
Shows that during tech revolutions, idiosyncratic uncertainty becomes systematic, producing "bubble" price patterns that are rational ex-ante. Explains why large alphas during tech booms are not necessarily evidence of irrational behavior.
Explosive Root Tests and PSY Bubble Detection (Magnitude Seven)
Basele & Phillips (2025)
Cowles Foundation Discussion Paper
Applies Phillips-Shi-Yu (PSY) bubble detection tests to the Magnificent Seven. Finds explosive root behavior in 6/7 stocks from Dec 2022 to Jan 2025. Key empirical motivation for our bubble-era framing.
AI Boom vs. TMT Bubble: An Investment Perspective
Amundi Investment Institute (April 2026)
Amundi Research White Paper
Directly compares AI (2023–2025) and TMT (1997–2000) rallies across concentration, valuation, and factor metrics. Finds AI most resembles the 1997–1999 "takeoff/exuberance" phase, not the 1999–2000 late-stage frenzy.
Is AI in a Bubble? (Goldman Sachs Equity Research, Issue 143)
Goldman Sachs Equity Research (2024)
Goldman Sachs Global Investment Research
Forward P/E comparison: AI stocks at 28× vs. dotcom leaders at ~50×. Provides valuation benchmarks for our cross-era comparison and contextualizes the smaller AI alpha under FF5.
Ken French Data Library — FF3 & FF5 Factor Files
French, K.R. (updated monthly)
Tuck School of Business, Dartmouth · mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html
Primary data source for all factor series. Files: F-F_Research_Data_Factors (FF3, monthly, 1963–present) and F-F_Research_Data_5_Factors_2x3 (FF5, monthly, 1963–present). Download format: YYYYMM CSV.
Sep 11, 2026 · Revised Proposal Due Sep 13 @ 5 PM

Team Meeting — Mustafa, Rinad & Arnav

Two days until the revised proposal deadline. This meeting is about dividing the five revision items, aligning on the dotcom comparison approach, and setting an internal handoff time.

Professor Feedback — Sep 3 Presentation (13/15)

1. No return plots in proposal. The proposal presentation had no data or visualizations. Professor wants return plots — stock excess returns (Y) vs. each FF factor (X). This cost 2 points on Data & Visualization.

2. Expand scope. Professor asked us to compare bubble eras directly — dotcom vs. AI side-by-side — rather than analyzing the AI period in isolation.

Task Division by Owner

Mustafa — Project Lead
  • 3
    Strengthen RMW hypothesis
    Write the formal H₀/H₁ framing — H₀: β_RMW = 0 across eras; H₁: β_RMW < 0 dotcom, > 0 AI. Research direction is his lane.
  • 4
    Add PSY bubble detection context
    Reference Basele & Phillips (2025) — 6/7 Mag-7 stocks show explosive root behavior Dec 2022–Jan 2025. This motivates the bubble-era framing in the intro.
  • 5
    Update literature section
    Add Jalali et al. (2026), Amundi Institute (April 2026), Goldman Sachs Issue 143. All strengthen the dotcom vs. AI framing.
Arnav — Econometric Analysis
  • 1
    Add return plots to revised proposal
    Proposal had no data at all — add scatter plots: returns (Y) vs. each FF factor (X). Four plots — Mkt-RF, HML, RMW, CMA. Already built on the site, need to go in the submitted doc.
  • 2
    Run dotcom-era FF3/FF5 regressions
    Produce the β estimates for the side-by-side factor loading comparison table. Mustafa can't finalize the RMW write-up without these numbers — deliver by tomorrow morning.
Rinad — Portfolio Construction
  • ✓
    Verify dotcom basket data alignment
    Confirm the 6-stock dotcom portfolio return data and time windows are correctly aligned for the extended dotcom-era regression. She owns the portfolio data.
  • ✓
    Formatting & proposal assembly
    Consolidate all revisions into the final revised proposal document for Canvas submission by Sep 13 at 5 PM.

Key Questions to Align on Today

1
How deep on the dotcom regressions? Are we re-running FF3/FF5 on the dotcom basket from scratch, or pulling estimates from existing literature (e.g. Ofek & Richardson 2003)? Affects how long item #2 takes.
2
Internal handoff deadline. Arnav needs to deliver dotcom β estimates before Mustafa can write the RMW hypothesis and comparison text. Agree on a time — ideally by tomorrow morning.
3
What goes in the revised proposal document vs. the site? The Canvas submission is a PDF/doc — make sure the factor loading table, RMW hypothesis, and literature additions all make it into the submitted version, not just the site.