{"paper_id":"2311.07478","title":"Optimal portfolio allocation with uncertain covariance matrix","authors":["Maxime Markov","Vladimir Markov"],"paper_date":"2023-11-13","paper_type":"theoretical","paper_url":"https://arxiv.org/pdf/2311.07478v1","page_url":"https://thequant.space/flowcharts/optimal-portfolio-allocation-with-uncertain-covariance-matri/","scores":{"empirical_rigor":4,"math_complexity":9,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The paper proposes a method that extends the mean‑variance approach by marginalizing the CARA expected utility over random covariance matrices and expected returns, yielding a logarithmic risk dependence and reduced allocation under higher uncertainty.","quote":"In this paper, we explore the portfolio allocation problem involving an uncertain covariance matrix. Our proposed method extends the mean-variance approach by considering the uncertainty associated with future covariance matrices and expected returns, which is important for practical applications.","summary":"In this paper, we explore the portfolio allocation problem involving an uncertain covariance matrix. We calculate the expected value of the Constant Absolute Risk Aversion (CARA) utility function, marginalized over a distribution of covariance matrices. We show that marginalization introduces a…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Portfolio returns x are assumed to follow a Gaussian distribution.","type":"model","stated":"explicit","quote":"P(x) can be assumed to follow a Gaussian distribution, as the fat tails of the actual return distribution can be attributed to the integration over variance.","support":"full"},{"assumption":"The future covariance matrix Σ and expected returns µ are treated as random variables with given probability distributions.","type":"statistical","stated":"explicit","quote":"If the future covariance matrix Σ and the expected returns µ are random variables, the optimal allocation weights are determined by integrating over all possible realizations of the covariance matrix Σ with probability distribution P (Σ) and the expected returns µ with probability distribution P(µ).","support":"full"},{"assumption":"The variance distribution is modeled by a shifted gamma distribution to capture minimal deterministic variance and stochastic part.","type":"statistical","stated":"explicit","quote":"As a result, we employ a gamma distribution shifted by minimal variance as the model for the future variance distribution. The shifted gamma distribution is ﬂexible enough to capture both ﬁnite sample uncertainty and forecasting power limitations by treating n as a model parameter.","support":"full"},{"assumption":"Integration over Σ involves all N(N+1)/2 independent components of the covariance matrix.","type":"model","stated":"explicit","quote":"The integration over Σ is a basic operation in multi-variate statistics and involves integrating over all N(N+1) 2 independent components of Σ.","support":"full"},{"assumption":"CARA utility function is used with risk‑aversion parameter a.","type":"model","stated":"explicit","quote":"We calculate the expected value of the Constant Absolute Risk Aversion (CARA) utility function, marginalized over a distribution of covariance matrices. d the risk aversion parameter a can be expressed as follows: Ua(x, a) = (1−exp(−ax) a , if a̸ = 0 x, if a = 0 , (1) To determine the optimal portfolio weights, we maximize the expected value of the utility function U(x, a) with respect to w: w = arg max w EP [Ua(x,","support":"full"},{"assumption":"Transaction costs can be incorporated as a quadratic turnover term proportional to η.","type":"execution","stated":"explicit","quote":"(30) which has the following classical MV solution: w∗= 1 aΣ−1µ (31) The eﬀect of the transaction cost can also be taken into account1. we obtain the expected utility after marginalization over random covariance matrices S: EΣ[U(µ, Σ)] ∼(−1)e−aµT w−α 2 ln h 1−a2 α (wT Σw) i (34) 1Additionally, one can take into account the transaction costs by including a term proportional to the turnover, e η 2 (w−w0)2, in the","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Simulated variances according to Eq. 11","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"Subsequently, N = 105 random variances were generated according to Eq. 11.","support":"full"},{"name":"Two‑dimensional Wishart matrix simulation","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"We simulate N = 105 samples and display the distribution of volatilities and the correlation coeﬃcient.","support":"full"},{"name":"VVIX Index (volatility of volatility)","provider":null,"kind":"market-prices","access":"public","period":null,"universe":null,"quote":"The volatility of volatility is measured by the VVIX Index, which represents the expected volatility (standard deviation) of the 30-day forward price of the VIX Index. The typical value of the VVIX index is in the 80 percent range.","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"unclear"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R4a","candidates":{"front_matter":"n/a","evidence_card":"partial","evidence_card_quoted":true},"card_detail":"A quadratic turnover term η·(w‑w0)^2 can be added to the utility, but costs are not included in the headline allocation results.","card_quote":"(30) which has the following classical MV solution: w∗= 1 aΣ−1µ (31) The eﬀect of the transaction cost can also be taken into account1. we obtain the expected utility after marginalization over random covariance matrices S: EΣ[U(µ, Σ)] ∼(−1)e−aµT w−α 2 ln h 1−a2 α (wT Σw) i (34) 1Additionally, one can take into account the transaction costs by including a term proportional to the turnover, e η 2 (w−w0)2, in the","meaning":"the paper reports no trading or execution result; the cost the extractor found is a model parameter"},"results":{"status":"available","evaluation_type":"none","findings":[{"finding":"Marginalization of the CARA utility over covariance uncertainty introduces a logarithmic dependence on risk instead of the linear dependence of the classic mean‑variance approach.","kind":"theoretical","quote":"We show that marginalization introduces a logarithmic dependence on risk, as opposed to the linear dependence assumed in the mean-variance approach.","support":"full"},{"finding":"Higher uncertainty in the covariance matrix leads to a decrease in the absolute level of portfolio allocation while preserving relative weights.","kind":"theoretical","quote":"Additionally, it leads to a decrease in the allocation level for higher uncertainties. In other words, accounting for uncertainty in the covariance results in a decrease in the absolute level while maintaining the same relative allocation.","support":"full"},{"finding":"Accounting for covariance uncertainty effectively shrinks the covariance matrix through marginalization of expected returns.","kind":"theoretical","quote":"In simple terms, the marginalization of expected returns results in the shrinkage of the covariance matrix.","support":"full"},{"finding":"The scaling function gW(q,α) satisfies 0 < gW ≤ 1, indicating that optimal weights are reduced by a factor depending on uncertainty parameter α.","kind":"theoretical","quote":"10 Figure 2: Scaling function gW (q, α) We note that the solution in Eq. 41 appears similar to the optimal weights calculated in [3] for outcomes that follow a Laplace distribution: w∗ LD = gLD(q) × \u00121 aΣ−1µ0 \u0013 (43) where gLD(q) = 1 −a2 ˜d 2 , ˜d = 2(1 + q −√1 + 2q) a2q (44) The scaling function 0 < gW (q, α) ≤1 does not dep","support":"full"},{"finding":"In the worst‑case scenario, the maximum weight is allocated to the worst‑performing asset, allowing an analytical solution via KKT conditions.","kind":"theoretical","quote":"Finally, we demonstrate that the worst-case scenario allows for an analytical solution using the Karush–Kuhn–Tucker (KKT) method [3].","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":24,"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":2,"extracted_at":"2026-11-14T20:00:41Z"},"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/"}