{"paper_id":"2309.17219","title":"Covariance matrix filtering and portfolio optimisation: the Average Oracle vs Non-Linear Shrinkage and all the variants of DCC-NLS","authors":["Christian Bongiorno","Damien Challet"],"paper_date":"2023-09-29","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2309.17219v1","page_url":"https://thequant.space/flowcharts/covariance-matrix-filtering-and-portfolio-optimisation-the/","scores":{"empirical_rigor":8.5,"math_complexity":6,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes the Average Oracle, a simple and very fast covariance filtering method, and evaluates it against DCC+NLS variants in large-scale randomized experiments.","quote":"The Average Oracle, a simple and very fast covariance filtering method, is shown to yield superior Sharpe ratios than the current state-of-the-art (and complex) methods, Dynamic Conditional Covariance coupled to Non-Linear Shrinkage (DCC+NLS).","summary":"The Average Oracle, a simple and very fast covariance filtering method, is shown to yield superior Sharpe ratios than the current state-of-the-art (and complex) methods, Dynamic Conditional Covariance coupled to Non-Linear Shrinkage (DCC+NLS). We pit all the known variants of DCC+NLS (quadratic…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The true covariance matrix changes only at fixed intervals of length T, remaining constant within each interval.","type":"model","stated":"explicit","quote":"Instead of assuming that C evolves slowly as a function of time as DCC+NLS does, we assume that it changes at fixed times kT, k ∈1, 2, · · ·, which also are opportunistically assumed to correspond to the end of successive estimation windows."},{"assumption":"At each time t the investment universe consists of the top N most capitalized US stocks that are listed throughout the in‑sample and out‑of‑sample windows.","type":"data","stated":"explicit","quote":"Consider day t an in-sample window [t −∆tin, t] and an out-of-sample window ]t, t + ∆tout]. The selection of available assets will be the top N most capitalized stocks of the US market on the day t such that all the stocks are listed in [t −∆tin, t + ∆tout];"},{"assumption":"Rebalancing is performed either weekly (Δt = 5 days) or monthly (Δt = 20 days).","type":"execution","stated":"explicit","quote":"We performed 10, 000 simulations by randomly sampling t and the initial set of stocks n available at time t. Tables 1 and 2 report performance metrics of long-short portfolios, for weekly (∆t = 5 days) and monthly ( ∆t =20 days) rebalancing."},{"assumption":"Transaction costs are a flat 5 basis points per trade, applied only at rebalancing.","type":"execution","stated":"explicit","quote":"To measure the respective performance of all the above methods, we set the transaction cost to 5 bps, which is considered a low transaction cost (Ledoit and Wolf, 2022a); larger transactions would increase the advantage of AO over DCCs."},{"assumption":"Stocks exiting the top‑1000 universe are sold and replaced by a random selection from the current top‑N universe.","type":"execution","stated":"explicit","quote":"At each rebalancing, if some stocks of the portfolio exit the universe N of the top 500 capitalized stocks, such stocks will be sold and substituted with a random selection from the universe N defined on the current day."},{"assumption":"The initial day t is drawn uniformly from 2000‑01‑03 to 2021‑01‑20, thus covering the 2008‑2009 and 2020 COVID crises.","type":"data","stated":"explicit","quote":"The initial day t is selected randomly in the range [2000-01-03,2021-01-20], which includes the 2008-2009 and 2020 COVID crises."}]},"datasets":{"status":"available","items":[{"name":"Top N most capitalized US stocks","provider":null,"kind":"market-prices","access":"unknown","period":"2000-01-03 to 2021-01-20 (random start d","universe":"Top N most capitalized US stocks (N≫1, e.g., top 500 or top 1000)","quote":"... The selection of available assets will be the top N most capitalized stocks of the US market on the day t such that all the stocks are listed in [t −∆tin, t + ∆tout]"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"Flat 5 basis points per trade, applied only at each rebalancing","quote":"To measure the respective performance of all the above methods, we set the transaction cost to 5 bps, which is considered a low transaction cost (Ledoit and Wolf, 2022a); larger transactions would increase the advantage of AO over DCCs.","source":"front_matter","rule":"R3","candidates":{"front_matter":"yes","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Flat 5 basis points per trade, applied only at each rebalancing","card_quote":"To measure the respective performance of all the above methods, we set the transaction cost to 5 bps, which is considered a low transaction cost (Ledoit and Wolf, 2022a); larger transactions would increase the advantage of AO over DCCs.","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"The Average Oracle yields higher Sharpe ratios than all DCC+NLS variants across large‑scale randomized experiments.","kind":"performance","quote":"The Average Oracle, a simple and very fast covariance filtering method, is shown to yield superior Sharpe ratios than the current state-of-the-art (and complex) methods, Dynamic Conditional Covariance coupled to Non-Linear Shrinkage (DCC+NLS)."},{"finding":"AO also achieves lower turnover, lower gross leverage, and higher diversification (effective number of assets) compared with the other methods.","kind":"performance","quote":"its Sharpe ratio is significantly larger, once again because of both better average returns and realized volatility (except for DCC-QIS)."},{"finding":"Some DCC+NLS variants attain the lowest average realized volatility, but their higher turnover and leverage reduce average returns and Sharpe ratios.","kind":"performance","quote":"while some variants of DCC+NLS sometimes yield the lowest average realized volatility, albeit with a small improvement, their excessive gross leverage and investment concentration, and their 10-time larger turnover contribute to smaller average portfolio returns, which mechanically result in smaller realized Sharpe rat"},{"finding":"When DCC‑QIS is constrained to match AO’s turnover, it can achieve the best Sharpe ratio for long‑short portfolios, albeit with higher realized volatility.","kind":"performance","quote":"While the unconstrained AO still beats all the DCC+NLS and NLS methods for long-short portfolios for all metrics except realized volatility, DCC‑QIS, once constrained by AO1200, does bring the best Sharpe ratio, owing to its smaller realized volatility."},{"finding":"For long‑only portfolios, AO outperforms other methods on all metrics, though the advantage is less pronounced than for long‑short portfolios.","kind":"performance","quote":"The advantage of AO for long-only portfolios (Table 6) is less impressive but still outperforms the other three methods in all metrics: its Sharpe ratio is significantly larger, once again because of both better average returns and realized volatility (except for DCC‑QIS)."}],"reported_performance":null,"out_of_sample":{"method":"walk-forward","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":39,"citations":{"cited_by_in_corpus":1,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"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":"2026-11-14T18:21:42Z"},"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/"}