Quantitative Risk Analysis: ESG Portfolios, EVT & Copulas in R (opens GitHub repository in new tab)
Compares a high-ESG Sustainable Portfolio (MSFT, PEP, OKE, JPM) against a Conventional
Portfolio (AAPL, COST, WMB, AXP) over 7,377 trading days, combining Minimum Variance
optimization, General & Non-Central Student-\(t\) MLE, Peaks-Over-Threshold \(\text{GPD}\) tail
modelling, and bivariate Copulas.
- Bulk & Tail Estimation: Selects 3-parameter General Student-\(t\) (\(\hat{\nu}_A=3.81, \hat{\nu}_B=3.50\)) via BIC and fits \(\text{GPD}(\xi,\beta)\) above \(u_A=38, u_B=42\) using custom Mean Excess Plots with \(95\%\) CIs.
- Asymptotic ESG Tail Insight: Proves Sustainable Portfolio A has a steeper GPD shape (\(\hat{\xi}_A=0.4097\) vs. \(\hat{\xi}_B=0.2647\)), causing \(\text{ES}_\alpha\) to accelerate as \(\alpha \to 99.9\%\).
- Copula Dependence: Student-\(t\) Copula (\(\hat{\rho}=0.5405, \hat{\nu}=3.49\)) outperforms Gaussian, Frank, and Clayton copulas by BIC (\(-3,051.69\)).
Quantitative Risk Analysis: ESG Portfolios, EVT & Copulas in R
Enterprise Risk & Extreme Value Modelling | R (890 Lines) • 33-Page Report • SP9 / CS2 / SP7
Executive Context & Three-Part Scope
This 33-page technical report and 890-line R pipeline investigates whether high-ESG ("sustainable") equity portfolios exhibit lower tail risk than conventional portfolios, alongside an institutional evaluation of Fannie Mae's Green Mortgage-Backed Securities (MBS) programme and climate risk propagation across insurance lines.
Key Empirical Findings & Actuarial Methodology
- Part I (Green MBS & Insurer Climate Risk): Evaluated Fannie Mae Green MBS prepayment risk, extension risk, and greenwashing/impact reporting risk, and mapped physical vs. transition climate risk across Property, Liability, Health, and Life insurers.
- Part II (Sector Clustering, Minimum Variance & Parametric Fitting): Clustered
S&P 500 Sector ETFs (
XLF, XLP, XLK, XLE) and partitioned 8 stocks by LSEG ESG scores into Portfolio A (Sustainable:MSFT, PEP, OKE, JPM) and Portfolio B (Normal:COST, AAPL, WMB, AXP) (\(V_0 = \text{£}1,000\), \(n=7,377\) days). Fitted Normal, General Student-\(t\), and Non-Central \(t\) via MLE & Pearson's \(\chi^2\) goodness-of-fit; General Student-\(t\) (\(\hat{\nu}_A=3.81, \hat{\nu}_B=3.50\)) won by BIC (\(-27,540.31\) and \(-29,061.47\)). - Part III (POT-GPD Tail Modelling & Hybrid Expected Shortfall): Built custom vectorized Mean Excess Plots with \(95\%\) CIs to select thresholds \(u_A = \text{£}38\) (\(\hat{\beta}_A=11.07, \hat{\xi}_A=0.4097\)) and \(u_B = \text{£}42\) (\(\hat{\beta}_B=16.35, \hat{\xi}_B=0.2647\)). Showed that Portfolio A's heavier tail index (\(\hat{\xi}_A = 0.4097\)) causes its Expected Shortfall to grow faster at extreme Solvency II confidence levels (\(\alpha \in [0.95, 0.999]\)).
- Multivariate Copula Dependence: Calibrated Gaussian, Student-\(t\), Clayton, and Frank copulas via Maximum Pseudo-Likelihood; the Student-\(t\) Copula (\(\hat{\rho}=0.5405, \hat{\nu}=3.49\)) achieved the best BIC (\(-3,051.69\)).