Buying the ten highest ESG scores in the Mexican large-cap index raised conditional volatility by 26% to 30% against the index itself in every year from 2020 to 2024. The paper gives no portfolio return, Sharpe ratio or realized P&L with which to price that extra risk.
The trade could hardly be simpler. Screen the S&P/BMV IPC, the headline index of the Mexican Stock Exchange, using Refinitiv ESG scores, then own the leaders. The attached claim says high-scoring companies should endure stress better and fund themselves more cheaply.
Paula Margarita Fosado Olvera and Roberto Yoan Castillo Dieguez, the two Querétaro authors, test the idea through four books. The High ESG book holds the top ten by score, while Low ESG holds the bottom ten. The benchmark is the index itself. A fourth, optimized ESG book maximizes the Sharpe ratio subject to ESG constraints. Its universe contains the 35 IPC constituents, which the paper says represent over 90% of Mexican market value. Only firms with complete annual statements and Refinitiv Eikon scores for 2020 to 2024 remain.
Each book then gets a GARCH(1,1) fit to daily log returns, with one fit for every portfolio and year. The authors also count occasions when realized volatility or drawdown crosses a limit. Kupiec and Christoffersen tests compare that hit sequence with a 5% expected violation rate. High ESG emerges with higher volatility and more breaches than the index. In the financing-cost model, governance and lagged financing cost explain WACC better than aggregate ESG.
The familiar resilience argument supplies the economic story, and the paper's descriptives appear to support it. High ESG companies are larger, better governed and less levered. Governance has the narrowest dispersion among the three pillars across the 35-firm panel, with SD 0.81 versus 2.44 for E and 2.17 for S. Size and low leverage usually accompany cheaper capital and lower realized volatility. Mean WACC, the weighted average cost of capital, is 10.59%, spanning 4.45% to 14.18%. That range leaves ample cross-sectional variation for sustainability scores to explain. The second stage tackles the question directly through a Random Forest predicting WACC from the ESG pillars, aggregate ESG, D/E, firm size, market capitalization and lagged WACC.
High scores brought higher volatility
High ESG records average conditional volatility of 1.38% in 2020, against 1.06% for the benchmark. By 2024, the figures are 1.35% and 1.07%. The smallest gap arrives in 2022, when High ESG posts 1.34% beside the benchmark's 1.06%, still a 26% difference. Low ESG, supposedly the penalized basket, stays close to the index at 1.08% to 1.14%.
The breach counts sharpen the result. High ESG exceeds its volatility limit 50.66% of the time on average, compared with 17.00% for the benchmark and 21.77% for Low ESG. The benchmark also has the lowest drawdown breach rate, at 29.08%.
Binomial tests reject adequate volatility control for all four books in 2020, 2022 and 2024. Only High ESG fails in 2021, with p = 0.0000. Low ESG at p = 0.2812, the benchmark at p = 0.9958 and the optimized book at p = 0.9384 show no evidence of violation during that year. High ESG and Optimized ESG both fail in 2023 at p = 0.0000. Low ESG at 0.9704 and the benchmark at 0.9878 clear the test.
Because the levels reject almost everything, the ranking contains the useful information. The passive index comes first. The authors acknowledge as much: "The Benchmark portfolio is positioned as the most stable, followed by the Low ESG portfolio." Their recommendation of the constructed books is therefore the result that needs explaining.
The volatility framing also leaves out a qualification. GARCH mean parameters provide the paper's only return-side figures, and those favour the screen. High ESG mu reaches 0.0495 in 2023 and 2024 against 0.0335 for the benchmark. In 2020, the comparison is 0.0380 against 0.0329. Low ESG has a negative 2020 mean parameter of -0.0047. Beyond these parameters, no portfolio return, Sharpe ratio or drawdown magnitude is reported. The screen clearly purchased volatility. Its portfolio-level return payoff remains unprinted.
Can optimization repay the constraint?
The abstract admits the central weakness: the optimized book "better contained volatility but struggled with extreme losses." Even at full strength, that concession fares poorly beside the benchmark column. Optimized volatility declines from 1.29% in 2020 to 1.18% in 2022, then reaches 1.23% in 2023 and 2024. It stays below High ESG every year and above the index's 1.06% to 1.07% throughout.
Its volatility breach rate is 35.26%, more than twice the benchmark's 17.00%. Drawdown breaches reach 42.61%, the highest rate among the four books and above High ESG's 36.01%.
The optimizer improved a book you would not have held, then lost to the book available for free.
The authors still defend the construction. Governance, leverage and lagged WACC lead their capital-cost prediction, a pattern they say supports "hybrid models like the Optimized ESG portfolio despite the persistent drawdown risk." Their conclusion assigns the book "relative advantages in volatility control and balance between risk and sustainability." Its volatility advantage exists against High ESG and disappears against the index. Drawdown risk belongs at the center of the design assessment. At 42.61%, the optimized book has the table's worst breach rate.
Persistence offers it a better comparison. Alpha plus beta totals 0.898 for the optimized book in 2023, below the benchmark's 0.914. High ESG requires a more guarded reading. The paper puts alpha plus beta near 0.97, while the components in its parameter table sum to 0.9671 in 2022 and 0.9748 in 2023. Low ESG reaches 0.9775 and 0.9771 in those years, exceeding High ESG both times. Every book sits close to a unit root on these estimates, with Low ESG highest.
Yesterday's WACC carries the model
The Random Forest contains the result a fundamentals shop can use. Aggregate ESG ranks last among eight features at 0.057, below firm size at 0.068. Governance alone ranks second at 0.210, followed by D/E in third place at 0.144. Combining the pillars erases the signal carried by governance. The authors connect this result to earlier Latin American evidence that identifies governance as the strongest factor reducing financing costs.
Lagged WACC leads all features at 0.257. Since WACC persists, predicting this year's financing cost from last year's largely captures its autocorrelation. The reported explained variance of 0.8912 and MAPE of 13.24% inherit that feature. According to the methodology, the model "is validated with cross-validation and grid search to ensure predictive accuracy and avoid overfitting." Yet the validation table reports an out-of-bag score of 0.1041. The paper never reconciles it with the same 0.8912, while describing the fit as strong performance. A gap that large has the appearance of an overfit forest.
Five years and a narrow universe
The authors identify several limitations themselves. The 35-firm universe leaves out medium-sized issuers, dependence on Refinitiv scores may create measurement bias, and sectoral or rolling ESG variants were "explored but omitted for feasibility and editorial conciseness." Some specifications therefore remain unreported.
Two further problems go unflagged. Requiring complete statements and ESG scores across 2020 to 2024 selects the universe using the full evaluation window in a 35-name market. The parameter tables also repeat rows. Every 2024 GARCH row is identical to its 2023 counterpart for all four portfolios, digit for digit, while the benchmark's 2023 row matches its 2021 row. The alternatives are duplicated fits or sixteen coincident numbers.
We could not locate either the threshold level or the estimation window used for the violation indicator. Numerical details for the optimization objective and constraint set are also absent. The backtesting procedure is said to use a rolling window for 2019 to 2024, whereas the published violation rates span 2020 to 2024. Profitability appears in the paper's question, yet none of the four books has a reported portfolio return or Sharpe ratio. The financing-cost leg has an actual model. The profitability leg has nothing.
We could not test any of this ourselves. Our equity coverage consists of US listings and ETFs. With no Mexican prices or ESG rating history, both the universe and sorting variable are missing rather than approximable.
The authors explain the result through Mexico's relatively shallow market, where ESG-labelled assets remain concentrated among few large issuers. On that reading, the score screen amounts to a concentration trade carrying a risk-reduction label. Governance at 0.210 contributes more to the financing-cost model than the composite score at 0.057. A version printing returns and weights beside the 1.23% volatility, with an optimized book that actually beats the benchmark column, would change my mind.