{"paper_id":"2507.01918","title":"End-to-End Large Portfolio Optimization for Variance Minimization with Neural Networks through Covariance Cleaning","authors":["Christian Bongiorno","Efstratios Manolakis","Rosario Nunzio Mantegna"],"paper_date":"2025-07-02","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2507.01918v2","page_url":"https://thequant.space/flowcharts/end-to-end-large-portfolio-optimization-for-variance-minimiz/","scores":{"empirical_rigor":8,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a rotation‑invariant neural network that directly outputs the global minimum‑variance portfolio by learning lag‑transformed returns and regularising covariance eigenvalues.","quote":"We develop a rotation-invariant neural network that provides the global minimum-variance portfolio by jointly learning how to lag-transform historical returns and marginal volatilities and how to regularise the eigenvalues of large equity covariance matrices.","summary":"We develop a rotation-invariant neural network that provides the global minimum-variance portfolio by jointly learning how to lag-transform historical returns and how to regularise both the eigenvalues and the marginal volatilities of large equity covariance matrices. This explicit mathematical…","strategy":"Neural Network global minimum-variance estimator"},"assumptions":{"status":"available","items":[{"assumption":"Each return vector is drawn from a multivariate normal distribution with zero mean and covariance Σ.","type":"statistical","stated":"explicit","quote":"assume each rt is drawn from a multivariate normal distribution with zero mean and covariance matrix Σ."},{"assumption":"A one‑day shift between training inputs and out‑of‑sample returns is used to avoid data leakage.","type":"data","stated":"explicit","quote":"OOS returns is introduced to avoid data leakage"},{"assumption":"Trades are executed at the actual close price without market impact.","type":"execution","stated":"explicit","quote":"The trades are executed with the actual close price of the day pt, without accounting for market impact."},{"assumption":"The calibration window is fixed to 1200 trading days and the number of trainable lag variables is 2Δtin.","type":"model","stated":"explicit","quote":"The number of days Δtin is fixed to 1200 trading days; in fact, the presence of the module of Sec. 3.1 includes 2Δtin trainable variables (αt, βt)"}]},"datasets":{"status":"available","items":[{"name":"US equities dataset","provider":null,"kind":"market-prices","access":"unknown","period":"January 1990 to December 2024","universe":"US equities","quote":"We trained our model on a dataset of US equities spanning January 1990 to December 2024."},{"name":"Top‑1,000 market‑cap stocks","provider":null,"kind":"market-prices","access":"unknown","period":null,"universe":"top n = 1,000 by market cap","quote":"we select the top n = 1, 000 for marketcap, defined from the previous day’s close price."}],"known":[],"sample":{"start":"2000","end":"2024","market":"US equities","asset_class":"equities","frequency":"daily"},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":true,"detail":"Commissions 0.035%/share below 300k shares, 0.020% above; exchange/clearing fees total 0.0845% of notional; SEC Section 31 fee 1.157 bps on sell notional.","quote":"First, commissions follow the IBKR tiered schedule for US equities: a per-share rate of 0.035% when the month-to-date routed volume is below 300,000 shares and 0.020% above that threshold, with a floor of $0.35 per ticket.","source":"front_matter","rule":"R3","candidates":{"front_matter":"yes","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Commissions 0.035%/share below 300k shares, 0.020% above; exchange/clearing fees total 0.0845% of notional; SEC Section 31 fee 1.157 bps on sell notional.","card_quote":"First, commissions follow the IBKR tiered schedule for US equities: a per-share rate of 0.035% when the month-to-date routed volume is below 300,000 shares and 0.020% above that threshold, with a floor of $0.35 per ticket.","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"The NN achieves the highest Sharpe ratio (1.011) with annualized volatility 10.9% and mean return 11.0% in unconstrained backtests, outperforming all other estimators.","kind":"performance","quote":"The unconstrained (long–short) results show that the NN achieves the highest Sharpe ratio (1.01) with an annualized volatility of 10.9% and a mean return of 11.0%."},{"finding":"The NN exhibits the highest turnover (112% every five trading days) and leverages 3.07, indicating higher trading intensity than competitors.","kind":"performance","quote":"The network rebalances more than its entire notional every five trading days (turnover 112.0%), and QIS rebalances 44.0% of its portfolio at each rebalance."},{"finding":"In realistic backtests with market‑order execution, slippage and fees, the NN still delivers lower realized volatility, smaller maximum drawdowns, and higher Sharpe ratios than state‑of‑the‑art competitors.","kind":"performance","quote":"In out-of-sample tests from January 2000 to December 2024, the estimator delivers systematically lower realized volatility, smaller maximum drawdowns, and higher Sharpe ratios than the best competitors, including state-of-the-art non-linear shrinkage"}],"reported_performance":{"strategy":"Neural Network global minimum-variance estimator","sharpe":1.011,"sharpe_benchmark":null,"annual_return_pct":11,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":35,"citations":{"cited_by_in_corpus":2,"by_independent_authors":1,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":1,"unclassified":1},"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":"2026-11-14T17:34:04Z"},"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/"}