{"paper_id":"2501.03938","title":"In-Sample and Out-of-Sample Sharpe Ratios for Linear Predictive Models","authors":["Antoine Jacquier","Johannes Muhle-Karbe","Joseph Mulligan"],"paper_date":"2025-01-07","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2501.03938v3","page_url":"https://thequant.space/flowcharts/in-sample-and-out-of-sample-sharpe-ratios-for-linear-predict/","scores":{"empirical_rigor":6,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper derives analytical approximations for the in‑sample and out‑of‑sample Sharpe ratios of portfolios built from linear predictive models.","quote":"This paper derives analytical approximations for the in-sample and out-of-sample Sharpe ratios of portfolios constructed using linear prediction models.","summary":"We study how much the in-sample performance of trading strategies based on linear predictive models is reduced out-of-sample due to overfitting. More specifically, we compute the in- and out-of-sample means and variances of the corresponding PnLs and use these to derive a closed-form approximation…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Signals and residuals are independent, iid across time and follow multivariate Gaussian distributions.","type":"statistical","stated":"explicit","quote":"We assume that the sequences {st} and {ϵt+1} are independent of each other, and iid across time with distributions s and e respectively, which are assumed to be multivariate Gaussian"},{"assumption":"Linear relationship between returns and signals with parameter β.","type":"model","stated":"explicit","quote":"The researcher assumes that the relationship between the signals and the returns follows a linear model with parameter β ∈Rm×p: rt+1 = βst + ϵt+1"},{"assumption":"For the main closed‑form result, signals are centred standardised (µs=0, Σs=Ip) and Z=Σ⁻¹_ε.","type":"model","stated":"explicit","quote":"we present here our main result for a special (but extremely common) case, taking Z = Σ⁻¹_ϵ"},{"assumption":"Ordinary least‑squares (OLS) estimator is used to fit β.","type":"statistical","stated":"explicit","quote":"this can be done using ordinary least-squares (OLS) by stacking the returns across all time steps into the matrix R"},{"assumption":"Training and test periods are non‑overlapping; out‑of‑sample period uses the same β estimate without re‑estimation.","type":"data","stated":"explicit","quote":"After estimating the predictive model on the training dataset, the researcher deploys the model out-of-sample at discrete trading times T2 = {T1 + 1, . . . , T1 + T2 + 1}"},{"assumption":"When simulating data, signals and residuals are generated as Gaussian iid or AR(1) with fat‑tails.","type":"other","stated":"explicit","quote":"We first test our approach in the case where the data‑generating model satisfies our key assumptions, namely that the signals and residuals are Gaussian and iid."}]},"datasets":{"status":"available","items":[{"name":"Commodity futures simulation case study","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":"commodity futures","quote":"The substantial quantitative importance of these effects is illustrated with a simulation case study for commodity futures following the methodology of Gârleanu and Pedersen"},{"name":"Goyal, Welch and Zafirov dataset","provider":"Goyal, Welch, Zafirov","kind":"macro","access":"public","period":null,"universe":null,"quote":"empirical case study using the dataset compiled by Goyal, Welch and Zafirov"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"unclear"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"unclear","evidence_card_quoted":false},"meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"The out‑of‑sample replication ratio falls as the number of assets or signals grows, especially when assets are correlated.","kind":"performance","quote":"Figure 7 shows that if these assets are completely independent, then adding additional assets results in both an increased replication ratio and out-of-sample Sharpe ratio. However, even with just 1% average correlation between assets we see that the replication ratio already becomes monotonically decreasing."},{"finding":"Panel regression dramatically improves replication ratio (70% vs 38%) and yields more consistent out‑of‑sample Sharpe ratios compared to the full model.","kind":"performance","quote":"the replication ratio is 70%, much higher than the 38% observed when the portfolio is constructed using the beta from the full regression"},{"finding":"Simulation under Gaussian iid assumptions matches analytical approximations for in‑sample and out‑of‑sample Sharpe ratios and replication ratio.","kind":"theoretical","quote":"In Figure 11 we see that the simulated expected in-sample and out-of-sample Sharpe ratios align perfectly with our analytical results, and so does the replication ratio."},{"finding":"Violating Gaussian iid assumptions (AR(1) signals, t‑distributed residuals) has little impact on the replication ratio, though both in‑sample and out‑of‑sample Sharpe ratios decrease.","kind":"performance","quote":"analytical replication ratio is still remarkably close to its simulation observed counterpart… Relaxing the Gaussian assumption to allow for fat tails causes the in-sample Sharpe ratio to decrease, but also causes the out-of-sample Sharpe ratio to decrease."},{"finding":"Increasing the length of the in‑sample period reduces overfitting risk and raises the out‑of‑sample replication ratio.","kind":"theoretical","quote":"Conversely, by extending the length of the in-sample period one can reduce the overfitting risk, and can produce a higher replication ratio out of sample."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":28,"citations":{"cited_by_in_corpus":1,"by_independent_authors":1,"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":"2026-11-14T17:48:13Z"},"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/"}