{"paper_id":"2411.18830","title":"Double Descent in Portfolio Optimization: Dance between Theoretical Sharpe Ratio and Estimation Accuracy","authors":["Yonghe Lu","Yanrong Yang","Terry Zhang"],"paper_date":"2024-11-28","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2411.18830v1","page_url":"https://thequant.space/flowcharts/double-descent-in-portfolio-optimization-dance-between-theo/","scores":{"empirical_rigor":3,"math_complexity":8.5,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The paper proposes that mean‑variance portfolio optimization exhibits a double‑descent‑like “double ascent” Sharpe ratio curve as model complexity (number of assets) increases.","quote":"we find that the performance of low-dimensional models initially improves with complexity but then declines due to overfitting. As model complexity becomes sufficiently high, the performance improves with complexity again, resulting in a “double ascent” Sharpe ratio curve similar to the “double descent” phenomenon obse","summary":"We study the relationship between model complexity and out-of-sample performance in the context of mean-variance portfolio optimization. Representing model complexity by the number of assets, we find that the performance of low-dimensional models initially improves with complexity but then declines…","strategy":"mean-variance portfolio optimization with varying asset count"},"assumptions":{"status":"available","items":[{"assumption":"High dimensionality: T, N → ∞ with ρ_T = N/T in (0,1)∪(1,∞).","type":"statistical","stated":"explicit","quote":"A1 (High Dimensionality). We consider asymptotic setup where T, N → ∞"},{"assumption":"Factorless model: excess returns rt = μ + y_t with i.i.d. zero‑mean, unit‑variance, bounded 4th‑moment y_t.","type":"model","stated":"explicit","quote":"A2 (Factorless Model). The excess returns vector at time t is generated as r_t = μ + y_t, where y_t is a vector of i.i.d. random variables with zero mean, unit variance and bounded 4th moment."},{"assumption":"Constant scale: the squared theoretical Sharpe ratio θ converges to a bounded limit ˜θ as N → ∞.","type":"model","stated":"explicit","quote":"A3 (Constant Scale). Assume that the square of theoretical Sharpe ratio, θ (equivalently"},{"assumption":"High dimensionality for factor model: same asymptotic regime with ρ_T in (0,1)∪(1,∞).","type":"statistical","stated":"explicit","quote":"B1 (High Dimensionality). We consider asymptotic setup where T, N → ∞"},{"assumption":"Factor model: excess returns rt = b μ_f + b σ_f ζ_t + σ_ε η_t with independent i.i.d. components having bounded 4th moments.","type":"model","stated":"explicit","quote":"η_t is a vector of i.i.d. random variables with zero mean, unit variance and bounded 4th moment, ζ_t"},{"assumption":"Constant scale for factor model: θ converges to ˜θ, implying ‖b‖^2 converges to a constant ˜b^2.","type":"model","stated":"explicit","quote":"B3 (Constant Scale). Assume that the square of theoretical Sharpe ratio, θ (equivalently"}]},"datasets":{"status":"available","items":[{"name":"CRSP monthly total individual equity returns","provider":"Center for Research in Security Prices (CRSP)","kind":"market-prices","access":"public","period":"January 1967 to June 2019 (630 months)","universe":"nearly 30,000 stocks, average >6,200 stocks per month","quote":"Our study utilizes a comprehensive dataset of monthly total individual equity returns sourced from the Center for Research in Security Prices (CRSP). This dataset includes firms listed on the New York Stock Exchange (NYSE), the American Stock Exchange (AMEX), and NASDAQ. The analysis spans from January 1967 to June 201","known_id":"crsp"}],"known":[{"id":"crsp","name":"CRSP","provider":"Center for Research in Security Prices","kind":"market-prices","access":"commercial","quote":"5.1 Data Our study utilizes a comprehensive dataset of monthly total individual equity returns sourced from the Center for Research in Security Prices (CRSP).","mentions":2,"used_in":1}],"sample":{"start":null,"end":null,"market":"US equity returns","asset_class":"equities","frequency":"monthly"},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":false,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"no","evidence_card_quoted":false},"meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"In high‑dimensional regimes (N > T) the out‑of‑sample Sharpe ratio ascends again, showing a double‑ascent pattern.","kind":"theoretical","quote":"What is important in this figure is the “ascent” observed in the high-dimensional region, when ρ > 1. From a finance perspective, the observed “ascent” suggests that increasing the number of assets beyond the sample size can enhance the out-of-sample risk-adjusted returns of the estimated portfolio."},{"finding":"Empirical results with CRSP data are consistent with the theory, showing the Sharpe ratio first rises with N then plateaus.","kind":"performance","quote":"Based on the empirical performance of portfolio models with different complexity, we can also estimate the relationship between the clairvoyant Sharpe ratio and the number of assets."},{"finding":"The calibrated clairvoyant Sharpe ratio follows θ(N)=θ_1 e^{−λ(N−1)}+¯θ(1−e^{−λ(N−1)}) and reaches half the upper bound with about 13 assets; at N≈39 the clairvoyant Sharpe is 1.11 (~¾ of the limit).","kind":"other","quote":"The speed parameter λ informs us how fast the clairvoyant Sharpe ratio increases with the number of assets. To reach half of the Sharp ratio upper bound, the portfolio needs to have around 13 assets. At the optimal low-dimensional point, when N = 39, the clairvoyant Sharpe ratio is 1.11, which is about three quarters o"},{"finding":"Theoretical Sharpe ratio upper bound would require about 12 independent asset‑pricing factors, aligning with other literature estimates.","kind":"theoretical","quote":"If we assume the asset pricing factors have the same Sharpe ratio as the market, then we need 12 independent asset pricing factors to reach the theoretical Sharpe ratio upper bound."}],"reported_performance":{"strategy":"mean-variance portfolio optimization with varying asset count","sharpe":1.4,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"walk-forward","headline_out_of_sample":true}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":26,"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":"2026-11-14T17:52:14Z"},"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/"}