{"paper_id":"2408.01772","title":"Investment strategies based on forecasts are (almost) useless","authors":["Michael Weba"],"paper_date":"2024-08-03","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2408.01772v1","page_url":"https://thequant.space/flowcharts/investment-strategies-based-on-forecasts-are-almost-useles/","scores":{"empirical_rigor":3,"math_complexity":7,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The paper proposes three optimal forecasts for future stock returns: a best measurable forecast, a best linear forecast, and a best linear unbiased forecast, with explicit formulas.","quote":"Then the best measurable forecast p∗ S, the best linear forecast epS and the best linear unbiased forecast bpS are given by p∗ S = pT + β (S −T), (5) epS = S + γ2 T + γ2 pT, (6) bpS = S T pT.","summary":"Several studies on portfolio construction reveal that sensible strategies essentially yield the same results as their nonsensical inverted counterparts; moreover, random portfolios managed by Malkiel’s dart-throwing monkey would outperform the cap-weighted benchmark index. Forecasting the future…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Asset price follows the Black‑Scholes dynamics Pt = P0 exp(α t + σ Wt) with a standard Wiener process.","type":"model","stated":"explicit","quote":"ck-Scholes model postulates that ﬂuctuations of an asset price Pt at time t follow the equation Pt = P0 exp(α t + σ Wt), t ≥0, (1) where P0 is the initial price, α stands for the trend coeﬃcient, σ > 0 de-notes volatility, and Wt, t ≥0, is a standard Wiener process with mean values E(Wt) = 0 and covariances E(Ws Wt) = min(s, t) for s, t ≥0.","support":"full"},{"assumption":"The increments of the price (or Poisson) process are independent.","type":"model","stated":"explicit","quote":"Essentially, only the independence of increments and properties of the ﬁrst and second moments are required whereas assumptions on distributions have nowhere been used.","support":"full"},{"assumption":"All parameters α, σ, λ, ν, τ are unknown, as is the trend coefficient β.","type":"statistical","stated":"explicit","quote":"All parameters α, σ, λ, ν, τ 2 are assumed to be unknown which is the usual situation encountered in practice.","support":"full"},{"assumption":"A sample path of returns is observed over a fixed interval [0,T] and the investor wishes to forecast at a future time S>T.","type":"data","stated":"explicit","quote":"Consider the following situation: a sample path of the price process has been observed over the time interval [0, T], and a potential investor wishes to esti-mate the future development of the price by means of the observed data; for instance, the investor may estimate the trend coeﬃcient and compute an ap-propriate forecast.","support":"full"},{"assumption":"The trend coefficient β is non‑zero (β≠0) for the main results.","type":"model","stated":"explicit","quote":"Suppose FT ⊂B and β̸ = 0.","support":"full"},{"assumption":"Only first‑ and second‑moment properties of the price process are required; distributional assumptions are not used.","type":"statistical","stated":"explicit","quote":"Essentially, only the independence of increments and properties of the ﬁrst and second moments are required whereas assumptions on distributions have nowhere been used.","support":"full"}]},"datasets":{"status":"none_found","items":[],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"none"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"none","findings":[{"finding":"The best measurable forecast has the smallest mean‑square error, followed by the best linear forecast, then the best linear unbiased forecast and finally the trivial forecast, with ordering depending on the relation between T and γ².","kind":"performance","quote":"One ﬁnds ∆(p∗ S) < ∆( epS) < min (∆( bpS), ∆(p◦ S)) with ∆( bpS) < ∆(p◦ S) for T > γ2, ∆( bpS) = ∆(p◦ S) for T = γ2 and ∆( bpS) > ∆(p◦ S) for T < γ2.","support":"full"},{"finding":"When the critical relation T < γ² holds, the best linear unbiased forecast is even worse than the trivial forecast.","kind":"negative","quote":"The main problem occurs if the critical relation T < γ2 is satisﬁed which implies that the best linear unbiased forecast is even worse than the trivial forecast.","support":"full"},{"finding":"Because β is unknown, the best measurable forecast p*ₛ = p_T + β(S−T) is practically worthless, and substituting any estimator for β cannot improve upon the best linear unbiased forecast.","kind":"theoretical","quote":"Due to the fact that β is unknown, however, p∗ S = pT + β(S −T) is practically worthless, and replacing β by a ’reasonable’ estimator does not remedy the situation. For ex-ample, if an arbitrary linear unbiased estimator is substituted for β then the resulting forecast cannot be better than bpS; furthermore, recall that an L2-consistent best linear unbiased estimator of β does not exist in general.","support":"full"},{"finding":"The optimal forecasts work in the presence of exogenous shocks and require only the last return p_T (i.e., the initial and final prices), making the interior price history irrelevant.","kind":"theoretical","quote":"Firstly, the advantages of the optimal forecasts are to be mentioned: they are working in the presence of exogeneous shocks, and they require only knowledge of the last return pT = log( PT/P0 ). The latter property entails that initial price P0 and ﬁnal price PT are already suﬃcient with regard to prediction.","support":"full"},{"finding":"The results are not limited to the Black‑Scholes model; they extend to any price process with independent increments and finite first‑ and second‑moments.","kind":"theoretical","quote":"An inspection of the proof of the above theorem shows, however, that its assertions - and hence the following conclusions - are by no means restricted to this special approach. Essentially, only the independence of increments and properties of the ﬁrst and second moments are required whereas assumptions on distributions have nowhere been used.","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":36,"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":"2026-11-14T19:56:12Z"},"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/"}