{"paper_id":"2508.14986","title":"Variable selection for minimum-variance portfolios","authors":["Guilherme V. Moura","André P. Santos","Hudson S. Torrent"],"paper_date":"2025-08-20","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2508.14986v1","page_url":"https://thequant.space/flowcharts/variable-selection-for-minimum-variance-portfolios/","scores":{"empirical_rigor":7,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes using machine learning variable selection to parameterize minimum-variance portfolio weights with a large set of firm-level characteristics and their transformations.","quote":"We parameterize minimum-variance portfolio weights as a function of a large pool of firm-level characteristics as well as their second-order and cross-product transformations, yielding a total of 4,610 predictors.","summary":"Machine learning (ML) methods have been successfully employed in identifying variables that can predict the equity premium of individual stocks. In this paper, we investigate if ML can also be helpful in selecting variables relevant for optimal portfolio choice. To address this question, we…","strategy":"L2-boosting minimum-variance portfolio"},"assumptions":{"status":"available","items":[{"assumption":"The study directly parameterizes portfolio weights as a function of lagged firm-level characteristics and their cross‑products and second‑order transformations.","type":"model","stated":"explicit","quote":"consistent with our interest in the portfolio choice problem, we focus on directly parameterizing the objective of interest - portfolio weights - as a function of lagged firm-level characteristics and their cross-products and second-order transformations"},{"assumption":"An equally weighted (EW) benchmark portfolio is used as the reference for the parametric portfolios.","type":"execution","stated":"explicit","quote":"The parametric portfolios are obtained by assuming an equally weighted (EW) benchmark portfolio."},{"assumption":"Transaction costs are modeled as a flat fee of 10 basis points per transaction.","type":"execution","stated":"explicit","quote":"Table 3 reports the annualized Sharpe ratios both before and after transaction costs of 10 basis points"},{"assumption":"Predictors are winsorized at the 1st and 99th cross‑sectional percentiles to mitigate extreme observations.","type":"data","stated":"explicit","quote":"we cross-sectionally winsorize each predictor at the 1st and 99th cross-sectional percentiles"},{"assumption":"Each predictor is standardized to have zero cross‑sectional mean and unit standard deviation.","type":"data","stated":"explicit","quote":"we also follow Brandt et al. (2009) and standardize each predictor so that its cross-sectional mean is zero and its standard deviation is one"},{"assumption":"Missing characteristic values are set to zero.","type":"data","stated":"explicit","quote":"Missing characteristic values are set to zero."}]},"datasets":{"status":"available","items":[{"name":"95 monthly firm characteristics (augmented to 4,610 predictors)","provider":"Dacheng Xiu homepage (downloaded from https://dachxiu.chicag","kind":"fundamentals","access":"public","period":"January 1970 to June 2019","universe":"15,701 firms (US equities)","quote":"We use the dataset of 95 monthly firm characteristics utilized in Gu et al. (2020). The database was downloaded from Dacheng Xiu’s homepage"},{"name":"CRSP stock returns","provider":"CRSP","kind":"market-prices","access":"commercial","period":"January 1970 to June 2019","universe":"US equities corresponding to the 15,701 firms","quote":"Stock returns are sourced from CRSP.","known_id":"crsp"}],"known":[{"id":"crsp","name":"CRSP","provider":"Center for Research in Security Prices","kind":"market-prices","access":"commercial","quote":"The database was downloaded from Dacheng Xiu’s homepage (https://dachxiu.chicagobooth. edu/download/datashare.zip).4 Stock returns are sourced from CRSP.","mentions":1,"used_in":1},{"id":"french-library","name":"Kenneth French Data Library","provider":"Kenneth R. French","kind":"market-prices","access":"public","quote":"Second, we implement the minimum variance parametric portfolios using the 10The risk free rate was obtained from Ken French’s data library web site.","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":"US stocks","asset_class":"equities","frequency":"monthly"},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"10 basis points per transaction (flat fee applied to portfolio turnover)","quote":"Table 3 reports the annualized Sharpe ratios both before and after transaction costs of 10 basis points","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"10 basis points per transaction (flat fee applied to portfolio turnover)","card_quote":"Table 3 reports the annualized Sharpe ratios both before and after transaction costs of 10 basis points","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Machine‑learning variable selection yields minimum‑variance portfolios with lower risk than traditional sparse specifications.","kind":"performance","quote":"minimum-variance portfolios achieve lower risk relative to sparse specifications commonly considered in the literature"},{"finding":"The L2‑boosting method delivers the highest Sharpe ratios (1.076 and 0.814) among all strategies, both before and after transaction costs.","kind":"performance","quote":"When transaction costs are taken into account, only the L2-boosting and ridge methods yield parametric portfolios with positive Sharpe ratios (1.076 and 0.814 when using the 95 characteristics)."},{"finding":"Ridge‑based portfolios have standard deviations close to the equally weighted benchmark, indicating limited benefit when the predictor set is vastly expanded.","kind":"performance","quote":"the standard deviation of the ridge-based portfolios with the augmented data set is close to that obtained with the EW and VW strategies."},{"finding":"Portfolio turnover varies markedly by method: L2‑boosting and Lasso around 1.5, while ridge‑based portfolios have turnover around 0.2.","kind":"performance","quote":"Lasso, elastic net, and L2-boosting achieve portfolio turnovers around 1.5, whereas ridge-based portfolios have a much lower turnover of around 0.2."},{"finding":"Adalasso and L2‑boosting outperform all benchmarks in maximum drawdown and VaR metrics.","kind":"performance","quote":"The Table shows that adalasso and L2-boosting methods outperformed all other benchmarks in terms of both maximum drawdown…"}],"reported_performance":{"strategy":"L2-boosting minimum-variance portfolio","sharpe":1.076,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":38.6},"out_of_sample":{"method":"walk-forward","headline_out_of_sample":true}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2407.08748","title":"Covariance Matrix Analysis for Optimal Portfolio Selection","paper_date":"2024-06-23","authors":["Lim Hao Shen Keith"],"empirical_rigor":6,"page_url":"https://thequant.space/flowcharts/covariance-matrix-analysis-for-optimal-portfolio-selection/"},"question":"Which minimum‑variance portfolio construction method yields the lowest out‑of‑sample turnover?","this_paper_finding":"Portfolio turnover varies markedly by method: L2‑boosting and Lasso around 1.5, while ridge‑based portfolios have turnover around 0.2.","other_paper_finding":"Ridge‑MVP and EN‑MVP generally exhibit lower out‑of‑sample turnover than the sample‑based and PCA‑based MVPs, indicating more stable portfolios.","other_paper_quote":"Similar to the out‑of‑sample variance presented in Table 10, it is notable that enhancements in an estimator’s condition number","note":"Both findings show ridge‑based portfolios have the lowest turnover (≈0.2), whereas other methods like L2‑boosting/Lasso are around 1.5 and sample/PCA‑based are higher.","confidence":0.529}],"contradicts":[],"qualifies":[],"compared_with":58,"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":"2026-11-14T17:29:48Z"},"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/"}