Celebrity Board Governance as a Signal of Leverage Discipline in Cyclical Sectors

Evidence from Construction and Real Estate — Pre and Post-COVID

---

Overview

This research notebook investigates whether firms in the Construction and Real Estate (C+RE) sectors whose boards exhibit high Celebrity Governance Scores (CGS) demonstrate measurably better leverage discipline and risk-adjusted returns compared to non-celebrity peers — and whether that relationship shifted after the COVID-19 shock of March 2020.

The study constructs a C+RE universe from Russell 3000 constituents, scores each firm on a composite three-dimension CGS, then applies a multi-method empirical battery: Fama-French three-factor regressions, Difference-in-Differences (DiD), Propensity Score Matching (PSM), and a leverage suppression-effect test. The key finding is that CGS functions as a governance signal of underlying leverage discipline — not a mediation chain — with the celebrity effect strengthening once leverage is controlled, confirming a suppression structure.

---

Research Design

Celebrity Governance Score (CGS)

Each firm is scored across three dimensions, ranked percentile-style within the universe:

Dimension	Proxy	Interpretation
Image (d_image)	Market capitalization rank	Visibility and institutional prominence
Influence (d_influence)	12-month momentum rank	Perceived performance leadership
Sovereignty (d_sov)	Revenue rank	Operational scale and market position

CGS = mean(d_image, d_influence, d_sov)

Firms in the top 25% of CGS (default cutoff: 0.75) are classified as Celebrity; the remainder as Non-Celebrity.

Study Windows

Period	Dates
Pre-COVID	2015-01-01 → 2020-02-28
Post-COVID	2020-03-01 → 2025-04-30

---

Methodology

The notebook runs six empirical steps, each producing a saved output:

Step 1  →  Rebuild C+RE Universe
Step 2  →  Build 3-Dimension CGS
Step 3  →  Fetch Returns + Fama-French Factors
Step 4  →  Portfolio Construction & Window Splits
Steps 5–10  →  Six Tables + Three Figures

Tables

#	Method	Key Variables
Table 1	Descriptive statistics (Welch t-test)	Market cap, beta, D/E, current ratio, Debt/EBITDA, CGS
Table 2	Fama-French 3-factor regression (HAC SEs)	Alpha, market beta, SMB, HML, Sharpe — 4 portfolio × period combinations
Table 3	Difference-in-Differences (cluster-robust SEs)	Treat × Post coefficient; basic model and FF3-controlled model
Table 4	Propensity Score Matching (1:1 nearest-neighbor)	Matched post-COVID cumulative return and FF3 alpha
Table 5	Leverage signal / suppression-effect analysis	Paths a, b, c, c′; Sobel test; suppression confirmation
Table 6	Robustness checks	Alternative CGS cutoffs (70%, 80%), placebo time splits (2018, 2019)

Figures

#	Content
Figure 1	Cumulative returns — Celebrity vs. Non-Celebrity vs. S&P 500 (2015–2025) with COVID-19 shock marker
Figure 2	CGS score distribution by celebrity classification
Figure 3	Leverage distributions — Panel A: Debt/Equity, Panel B: Debt/EBITDA (Welch p-values from raw data; 95th-percentile cap for display only)

---

Repository Structure

.
├── cyclical_sector_celeb_bod_indictor.ipynb   # Main research notebook
├── README.md
└── outputs/                                   # Generated on run (Google Drive)
    ├── cre_universe.csv
    ├── cre_cgs.csv
    ├── table1_descriptive.csv
    ├── table2_ff3.csv
    ├── table3_did.csv
    ├── table4_psm.csv
    ├── table5_leverage_signal.csv
    ├── table6_robustness.csv
    ├── figure1_cumulative_returns.png
    ├── figure2_cgs_distribution.png
    └── figure3_leverage_distribution.png

---

Requirements

pip install yfinance pandas numpy statsmodels scikit-learn scipy matplotlib pandas-datareader

Package	Purpose
yfinance	Price history, fundamental data, ETF holdings
pandas_datareader	Fama-French factor download (Ken French data library)
statsmodels	OLS, FF3 regressions, DiD with clustered SEs
scikit-learn	Logistic regression (propensity scores), nearest-neighbor matching
scipy	Welch t-tests, Sobel test
matplotlib	All figures

Platform: Designed for Google Colab with Google Drive mounted at /content/drive/MyDrive/Russell3000_Diagnostic/. Adapt the DRIVE_DIR / OUT paths to run locally.

---

Quick Start

1. Open in Colab — upload or link the notebook, then mount your Drive:

   from google.colab import drive
   drive.mount("/content/drive")

2. Set output path — edit DRIVE_DIR in the first cell if needed:

   DRIVE_DIR = "/content/drive/MyDrive/Russell3000_Diagnostic"

3. Run all cells — the notebook is sequential; all steps must run in order for the panel dataset and portfolio returns to be available downstream.

4. Standalone figure regeneration — the final cells are self-contained and can be re-run independently to regenerate corrected Figure 1 (S&P 500 legend fix) and Figure 3 (Welch p-values from raw, uncapped data) without re-executing the full pipeline.

---

Key Parameters

PRE_START  = "2015-01-01"
PRE_END    = "2020-02-28"
POST_START = "2020-03-01"
POST_END   = "2025-04-30"
CGS_CUTOFF = 0.75          # Top 25% → Celebrity classification
MIN_MCAP   = 100e6         # $100M minimum market cap filter

---

Interpretation Notes

• Suppression, not mediation. Table 5 tests whether celebrity governance mediates returns through leverage reduction. It does not. When leverage is controlled, the celebrity coefficient increases — a textbook suppression effect. This means CGS and low leverage are independent predictors of post-COVID returns; CGS signals governance quality, which in turn is correlated with but not causally upstream of leverage discipline.
• Sobel test non-significance confirms the absence of a mediation chain and is consistent with suppression.
• Propensity score matching controls for pre-existing differences in size, beta, and leverage structure, isolating the governance signal effect.
• Placebo splits (Table 6) use 2018 and 2019 as alternative treatment thresholds to verify the pre-COVID period does not spuriously generate the same effect.

---

Data Sources

Source	Data
Yahoo Finance (yfinance)	Equity prices, market cap, D/E, beta, EBITDA, total debt
Sector ETFs (ITB, XHB, PKB, VNQ, IYR, XLRE, REM, REZ)	Initial universe construction
Kenneth French Data Library	Fama-French 3-factor monthly returns (Mkt-RF, SMB, HML, RF)