{"paper_id":"2506.23619","title":"Overparametrized models with posterior drift","authors":["Guillaume Coqueret","Martial Laguerre"],"paper_date":"2025-06-30","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2506.23619v1","page_url":"https://thequant.space/flowcharts/overparametrized-models-with-posterior-drift/","scores":{"empirical_rigor":6,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper extends misspecified isotropic results to non‑i.i.d. data and derives the expected return of a market‑timing strategy accounting for posterior drift.","quote":"First, in Section 2, we extend the misspecified isotropic results of Hastie et al. (2022) to the case of non-i.i.d. data and derive the expected return of a market-timing strategy when posterior drift is taken into account.","summary":"This paper investigates the impact of posterior drift on out-of-sample forecasting accuracy in overparametrized machine learning models. We document the loss in performance when the loadings of the data generating process change between the training and testing samples. This matters crucially in…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The empirical spectral distribution ˆµ converges weakly to a density measure µ with support on [0,∞).","type":"statistical","stated":"explicit","quote":"Assumption E.3. ˆµ weakly converges to a density measure µ. Besides, sup(µ)"},{"assumption":"The drift‑related measures ˆζis, ˆζoos and ˆζr converge weakly to limiting density measures.","type":"statistical","stated":"explicit","quote":"Assumption E.4. ˆζis, ˆζoos and ˆζr weakly converges to some density measures ζis, ζoos and ζr respectively."},{"assumption":"The model complexity is bounded away from the interpolation threshold (c ≠ 1).","type":"model","stated":"explicit","quote":"Assumption E.5. |c −1| ≥τ. The complexity of the model is supposed to be bounded away from the interpolation threshold, c = 1."},{"assumption":"The minimum eigenvalue of the covariance matrix is strictly positive (λp > τ).","type":"model","stated":"explicit","quote":"> τ. This assumption imposes positive-definiteness to Σ by requiring the minimum eigen- value of the covariance matrix to be strictly positive."}]},"datasets":{"status":"available","items":[{"name":"CRSP value‑weighted returns minus the T‑bill rate","provider":"CRSP","kind":"market-prices","access":"public","period":null,"universe":null,"quote":"We use the CRSP value-weighted returns minus the T-bill rate as dependent variable.","known_id":"crsp"},{"name":"14 macro‑economic indicators from Goyal et al. (2024) plus lagged returns","provider":"Amit Goyal’s website","kind":"macro","access":"public","period":null,"universe":null,"quote":"The predictors consist of the 14 macro-economic indicators used in Kelly et al. (2024), to which we add lagged returns."}],"known":[{"id":"crsp","name":"CRSP","provider":"Center for Research in Security Prices","kind":"market-prices","access":"commercial","quote":"This generates time-series of loadings ˆβt that characterize the predictive link between the predictors and the returns from the CRSP portfolio rt+1.","mentions":2,"used_in":2}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":null,"detail":null,"quote":"This is even before taking trading costs into account: holding the market portfolio entails minimum trading costs, which is not the case of the timing strategy presented in Kelly et al. (2024).","source":"front_matter","rule":"R3","candidates":{"front_matter":"no","evidence_card":"no","evidence_card_quoted":true},"card_detail":null,"card_quote":"This is even before taking trading costs into account: holding the market portfolio entails minimum trading costs, which is not the case of the timing strategy presented in Kelly et al. (2024).","meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Posterior drift leads to a loss in out‑of‑sample forecasting performance for over‑parametrized models.","kind":"predictive-accuracy","quote":"We document the loss in performance when the loadings of the data generating process change between the training and testing samples."},{"finding":"Holding periods of 15 years produce very heterogeneous returns for small bandwidths, while large bandwidths yield more consistent but less attractive risk‑adjusted returns.","kind":"performance","quote":"For the average investor, we find that focusing on holding periods of 15 years can generate very heterogeneous returns, especially for small bandwidths. Large bandwidths yield much more consistent outcomes, but are far less appealing from a risk-adjusted return standpoint."},{"finding":"Maximum Sharpe ratio observed across sub‑periods is about 0.3, lower than the ~0.4 reported in prior work.","kind":"performance","quote":"Over the full sample, we find a maximum value of 0.3, slightly lower than"},{"finding":"The market‑timing strategy’s Sharpe ratio is sensitive to both sub‑period selection and the kernel bandwidth parameter.","kind":"performance","quote":"Figure 8: Sharpe ratio sensitivity to periods and bandwidth. We plot average annualized Sharpe ratios obtained on six sub-periods and on the full sample for p = 600 RFFs, as a function of the bandwidth parameter of the kernel from the Random Fourier Features."},{"finding":"Theoretical results predict a natural degradation of performance as the shift between in‑sample and out‑of‑sample betas increases.","kind":"theoretical","quote":"Our theoretical results exhaustively and invariably point to a natural degradation of performance when the shift between in-sample and out-of-sample betas increases."}],"reported_performance":null,"out_of_sample":{"method":"walk-forward","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":42,"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":"2026-11-14T17:34:18Z"},"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/"}