{"paper_id":"2508.18679","title":"Identifying Risk Variables From ESG Raw Data Using A Hierarchical Variable Selection Algorithm","authors":["Zhi Chen","Zachary Feinstein","Ionut Florescu"],"paper_date":"2025-08-26","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2508.18679v1","page_url":"https://thequant.space/flowcharts/identifying-risk-variables-from-esg-raw-data-using-a-hierarc/","scores":{"empirical_rigor":7,"math_complexity":6.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a Hierarchical Variable Selection (HVS) framework designed for ESG datasets with a hierarchical structure and many more variables than observations.","quote":"We propose a framework specifically designed for ESG datasets characterized by a hierarchical data structure and a significantly larger number of variables than observations.","summary":"Environmental, Social, and Governance (ESG) factors aim to provide non-financial insights into corporations. In this study, we investigate whether we can extract relevant ESG variables to assess corporate risk, as measured by logarithmic volatility. We propose a novel Hierarchical Variable…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Logarithmic transformation of volatility satisfies model assumptions and is used as the response variable.","type":"statistical","stated":"explicit","quote":"The results suggest that a logarithmic transformation (λ = 0) of volatility satisfies the model assumptions, as illustrated in Table 1."},{"assumption":"Missing values for boolean variables are set to zero and numerical variables with less than 80% coverage are removed.","type":"data","stated":"explicit","quote":"Many of the raw variables in the dataset contain missing values. We follow Cheng et al. (2014); Clarkson et al. (2008) by assigning 0 to missing values for boolean variables. We then remove numerical variables that have less than 80% data available."},{"assumption":"Observations with any remaining missing values after processing are excluded from the analysis.","type":"data","stated":"explicit","quote":"Finally, any company-year observation that still contains missing values after these steps is removed."},{"assumption":"The residuals of the regression are heteroskedastic and right‑skewed, requiring a transformation.","type":"statistical","stated":"explicit","quote":"Consistent with these economic phenomena, our diagnostic checks reveal that the model’s residuals are heteroskedastic and the distribution is right‑skewed."},{"assumption":"The hierarchical structure of ESG data (tiers of raw variables, category scores, pillar scores, overall score) is appropriate for partitioning variables in the selection process.","type":"model","stated":"explicit","quote":"LSEG uses the category level (Tier 2 in Figure 1) in order to partition raw variables."}]},"datasets":{"status":"available","items":[{"name":"LSEG ESG Dataset","provider":"LSEG (formerly Refinitiv)","kind":"alternative","access":"commercial","period":null,"universe":"US companies across multiple sectors","quote":"Taking the Energy sector as an example, the initial dataset has 617 variables and 695 company‑year data points.","known_id":"refinitiv"}],"known":[{"id":"refinitiv","name":"Refinitiv / LSEG (Datastream, Eikon, Tick History)","provider":"LSEG","kind":"market-prices","access":"commercial","quote":"The dashed lines indicate the Adjusted R-squared values obtained from linear regression models using aggregated scores provided by LSEG as regressors (i.e., 10 category scores, 3 pillar scores, single overall score). this Adjusted R-squared value is also larger than those reported in other studies (En-gelhardt et al. 2021; L¨o¨of et al. 2022; Zhou and Zhou 2021).","mentions":16,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"Raw ESG variables selected by HVS are significantly more relevant to financial risk than aggregated ESG scores.","kind":"statistical","quote":"We show that raw variables selected by the proposed framework are significantly more relevant to financial risk than aggregated ESG scores."},{"finding":"Using logarithmic volatility improves model fit, increasing R‑squared compared with regular volatility.","kind":"statistical","quote":"Using logarithmic volatility also leads to an increase in the R-squared value of our framework, indicating improved model fit and explanatory power."},{"finding":"HVS outperforms stepwise regression out‑of‑sample, showing lower MSE and stronger robustness.","kind":"performance","quote":"These results indicate that the HVS framework reliably identifies relevant variables while keeping strong robustness on out-of-sample data."},{"finding":"Selected raw ESG variables provide new information beyond traditional financial risk factors.","kind":"predictive-accuracy","quote":"The selected raw ESG variables bring new information to traditional financial risk factors."},{"finding":"Relevant ESG risk factors differ across sectors and between large‑ and small‑cap firms.","kind":"other","quote":"different sectors have different relevance to ESG risk. Furthermore, our results reveal that the relevant ESG risk factors differ between large- and small-cap companies."}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2309.05866","title":"An Axiomatic Risk-Reward Framework for Sustainable Investing","paper_date":"2023-09-11","authors":["Gabriele Torri","Rosella Giacometti","Darinka Dentcheva","Svetlozar T. Rachev","W. Brent Lindquist"],"empirical_rigor":5.5,"page_url":"https://thequant.space/flowcharts/an-axiomatic-risk-reward-framework-for-sustainable-investing/"},"question":"Do ESG metrics predict financial performance and risk?","this_paper_finding":"Raw ESG variables selected by HVS are significantly more relevant to financial risk than aggregated ESG scores.","other_paper_finding":"Aggregated data show a limited relationship between ESG scores and financial performance (returns and volatility).","other_paper_quote":"Panels C and D report the average ESG score across the entire period in relation to the average returns and the return volatility, respectively; looking at this aggregate data, the relation between the ESG score and the financial performance seems limited.","note":"Both find a limited relationship for aggregated ESG scores, while raw ESG variables are significantly more relevant to financial risk.","confidence":0.643}],"contradicts":[],"qualifies":[],"compared_with":52,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T17:28:58Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}