{"paper_id":"2311.01985","title":"Maximizing Portfolio Predictability with Machine Learning","authors":["Michael Pinelis","David Ruppert"],"paper_date":"2023-11-03","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2311.01985v1","page_url":"https://thequant.space/flowcharts/maximizing-portfolio-predictability-with-machine-learning/","scores":{"empirical_rigor":6,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes constructing a maximally predictable portfolio (MPP) of stocks using machine learning forecasts and applying a Kelly‑style timing strategy to achieve outperformance.","quote":"We construct the maximally predictable portfolio (MPP) of stocks using machine learning.","summary":"We construct the maximally predictable portfolio (MPP) of stocks using machine learning. Solving for the optimal constrained weights in the multi-asset MPP gives portfolios with a high monthly coefficient of determination, given the sample covariance matrix of predicted return errors from a machine…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Rt is a jointly stationary and ergodic stochastic process with finite expectation.","type":"statistical","stated":"explicit","quote":"Assumption 1 Rt is a jointly stationary and ergodic stochastic process with finite expectation E[Rt] = µ = [µ1, µ2, ..., µn]⊤and autocovariance matrices E[(Rt−k−µ)(Rt−µ)⊤] = Σk, where k ≥0 without loss of generality."},{"assumption":"The conditional forecast ˆZt equals the conditional expectation E[Zt|Ft−1].","type":"model","stated":"explicit","quote":"the conditional expectation of the asset returns with the information set at the previous time."},{"assumption":"Forecast errors ϵt are independent of the forecasts ˆZt.","type":"statistical","stated":"explicit","quote":"where E[ϵt|Ft−1] = 0 ∈IRn, var[ϵt|Ft−1] = Γ ∈IRn×n and importantly for the following results ˆZt and ϵt are assumed independent."},{"assumption":"Equation 3 holds only if forecast errors are independent from the forecasts.","type":"model","stated":"explicit","quote":"Equation 3 only holds if assuming the forecast errors are independent from the forecasts, which is often not realistic."},{"assumption":"Sample covariance matrices Σ0 and Γ are estimated each month with rolling‑window historical data.","type":"data","stated":"explicit","quote":"In the empirical implementation, each expression above can be estimated with historical data Σ0 and Γ are estimated each month with rolling window sample data as sample covariance matrices."},{"assumption":"The information set Ft−1 includes observable economic variables such as interest‑rate spreads and stock‑specific factors.","type":"other","stated":"explicit","quote":"In the information set Ft−1 are observable leading economic variables like interest-rate spreads or stock-specific factors like momentum, fundamental company performance measures, liquidity, and volatility."}]},"datasets":{"status":"available","items":[{"name":"CRSP database","provider":"CRSP","kind":"market-prices","access":"public","period":null,"universe":null,"quote":"Stock prices, volume, and shares outstanding are from the CRSP database.","known_id":"crsp"},{"name":"Compustat database","provider":"Compustat","kind":"fundamentals","access":"public","period":null,"universe":null,"quote":"Fundamental company data is from Compustat.","known_id":"compustat"},{"name":"Kenneth French data library","provider":"Kenneth French","kind":"benchmark","access":"public","period":null,"universe":null,"quote":"Kenneth French’s website contains the historical risk‑free","known_id":"french-library"},{"name":"CRSP/Compustat merged database","provider":"CRSP/Compustat","kind":"market-prices","access":"public","period":null,"universe":"S&P 500 constituents","quote":"The CRSP/Compustat merged database gives stock returns and the set of stocks that are in the S&P 500 index each month.","known_id":"crsp"}],"known":[{"id":"compustat","name":"Compustat","provider":"S&P Global","kind":"fundamentals","access":"commercial","quote":"4 Empirical Results 4.1 Data This paper uses monthly data from CRSP, Compustat, and Kenneth French’s4 website.","mentions":3,"used_in":2},{"id":"crsp","name":"CRSP","provider":"Center for Research in Security Prices","kind":"market-prices","access":"commercial","quote":"4 Empirical Results 4.1 Data This paper uses monthly data from CRSP, Compustat, and Kenneth French’s4 website.","mentions":3,"used_in":1}],"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":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":"The MPPs achieve high monthly coefficients of determination (R²), with OLS attaining the highest average R² among the models.","kind":"predictive-accuracy","quote":"The predictability for portfolios formed with the different models is similar, but OLS actually attains the highest coefficient of determination over the three different weight maxes."},{"finding":"Machine‑learning‑based MPPs generate higher annual returns and Sharpe ratios than the S&P 500 benchmark.","kind":"performance","quote":"The machine learning MPP has higher annual returns than the market index."},{"finding":"Timing the MPP using its conditional excess‑return and volatility expectations improves risk‑adjusted returns relative to both the index and a reward‑risk‑timed index.","kind":"performance","quote":"When an investor times the machine learning MPP with its conditional excess return and volatility expectations, he or she can earn substantial improvements in risk‑adjusted returns over not just the index but the portfolio which reward‑risk times the index."},{"finding":"The Kelly‑style weighting scheme, where weights are proportional to the probability of a successful bet, is proposed to enhance out‑of‑sample performance.","kind":"theoretical","quote":"…a Kelly strategy (Kelly, 1956) where the MPP weights are proportional to the probability of successful bet."},{"finding":"Out‑of‑sample R² for individual stock return forecasts is reported, indicating the predictive power of the ML models.","kind":"predictive-accuracy","quote":"To assess the out‑of‑sample predictive performance for individual stock return forecasts, we calculate the out‑of‑sample R2 as R2 oos = 1 −"}],"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":1,"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":"2026-11-14T18:20:08Z"},"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/"}