Relevance
7/10
Highly relevant for quantitative portfolio managers and asset allocators seeking to incorporate physical climate risk into systematic investment strategies. The multi-objective optimization framework with MOPSO is directly applicable to portfolio construction. The backtesting period (2020-2025) includes major market events (COVID-19, 2022 energy shock), providing practical validation. However, the monthly rebalancing frequency and continent-level granularity make it more suited for strategic asset allocation than high-frequency trading. The novel CEV metric and its decomposition into idiosyncratic and systemic components offer actionable insights for risk-aware portfolio construction.
Implementation Complexity
7/10
The paper involves multiple complex components: (1) constructing standardized temperature anomaly time series with logistic regression probability estimation, (2) computing Fréchet-Hoeffding bounds for Bernoulli correlations, (3) defining climate-normalized portfolio weights incorporating asset intensity and revenue geography, (4) implementing a 3-objective MOPSO algorithm with Ledoit-Wolf shrinkage for covariance estimation, and (5) conducting rolling-window backtesting. The MOPSO implementation requires careful parameter tuning (population size, repository size, grid system, mutation rate). The optimization runs at ~217 seconds per month for 120 assets, which is computationally intensive but feasible. Data requirements include temperature records, MSCI constituent data, company-level revenue by continent, and tangible asset values.
Reproducibility
3/5
The paper uses publicly available temperature data from Our World in Data (sourced from IPCC, NASA, World Bank) and MSCI World index constituents. The MOPSO algorithm is well-documented with full parameter specifications in Appendix C. However, no code repository is explicitly provided, and the specific MSCI World constituent selection criteria (120 stocks via stratified sampling) may be difficult to replicate exactly. The logistic regression and panel regression specifications are fully detailed.