{"paper_id":"2407.08953","title":"Attribution Methods in Asset Pricing: Do They Account for Risk?","authors":["Dangxing Chen","Yuan Gao"],"paper_date":"2024-07-12","paper_type":"theoretical","paper_url":"https://arxiv.org/pdf/2407.08953v1","page_url":"https://thequant.space/flowcharts/attribution-methods-in-asset-pricing-do-they-account-for-ri/","scores":{"empirical_rigor":6,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes several axioms for reflecting risk in asset pricing and provides a thorough analysis of Shapley value (BShap) and Integrated Gradients, outlining their advantages and disadvantages.","quote":"Our work has made the following contributions: • Several axioms are proposed based on different aspects for reﬂecting risk in asset pricing. • Our examples demonstrate the motivations and signiﬁ-cance of these axioms. • We provide a thorough analysis of Shapley value and IG, outlining their advantages and disadvantages. II.","summary":"Over the past few decades, machine learning models have been extremely successful. As a result of axiomatic attribution methods, feature contributions have been explained more clearly and rigorously. There are, however, few studies that have examined domain knowledge in conjunction with the axioms.…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The feature space consists of n features.","type":"model","stated":"explicit","quote":"PRELIMINARIES For problem setup, assume we have n features.","support":"full"},{"assumption":"The class of functions F is the set of real analytic functions.","type":"model","stated":"explicit","quote":"We denote a class of functions f : Rn →R by F. We assume F to be the set of real analytic functions.","support":"full"},{"assumption":"Baseline point x′ is less than or equal to explicand x component‑wise (x ≥ x′).","type":"model","stated":"explicit","quote":"Without loss of generality (WLOG), we assume x ≥x′.","support":"full"},{"assumption":"The training domain X is convex for the Convex Geometry Axiom.","type":"data","stated":"explicit","quote":"We say a BAM preserves convex geometry (CG) if X is convex, x, x′ ∈X, then the calculation of A(x, x′, f) only requires f(X).","support":"full"},{"assumption":"Option pricing features are limited to underlying price, risk‑free rate, time to expiration, strike, and volatility.","type":"data","stated":"explicit","quote":"Models are based on ﬁve features, namely underlying stock (index) prices S, risk-free interest rates r, time to expiration τ, strike prices K, and volatility σ such that x = (S, r, τ, K, σ).","support":"full"},{"assumption":"Data are split randomly into 75 % training and 25 % test sets.","type":"execution","stated":"explicit","quote":"We randomly split data into 75% training data and 25% test data.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"2008 U.S. European call and put option data","provider":"Wharton Research Data Services","kind":"options","access":"proprietary","period":"2008","universe":"U.S. equity options","quote":"D. Data and Neural Networks Setup for Option Pricing We collected call and option data from Wharton Research Data Services in 2008. There are 253 trading days with 123969 records of option data in total.","support":"full","known_id":"wrds"},{"name":"VIX index (30‑day expected volatility of S&P 500)","provider":null,"kind":"macro","access":"public","period":null,"universe":null,"quote":"B. VIX VIX is the ticker symbol for the Chicago Board Options Exchange’s CBOE Volatility Index. The VIX we use is the 30-day expected volatility of the S&P 500 index, more precisely the square root of a 30-day expected realized variance of the index.","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"unclear"},"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":"unclear","findings":[{"finding":"BShap preserves average individual monotonicity, demand individual monotonicity, (reverse) diminishing marginal effect, increasing marginal effect, and first‑order monotonic dominance.","kind":"theoretical","quote":"BShap preserves average individual monotonic-ity (AIM), demand individual monotonicity (DIM), (reverse) diminishing marginal effect ((R)DME), increasing marginal effect (IME), and ﬁrst-order monotonic dominance (FMD).","support":"full"},{"finding":"Empirical results on option pricing during the 2008 crisis validate that both BShap and IG preserve risk patterns such as Delta sign, Gamma convexity, and Rho direction.","kind":"other","quote":"The results validated the Theorem V.1, V.2, and -100 -50 0 50 100 150 200 250 Attributions K t r S Features 2.61 189.57 -46.03 -4.65 -80.68 Fig. 2: IG for Eq 13 -100 -50 0 50 100 150 200 250 Attributions K t r S Features 1.96 183.4 -41.52 -8.5 -74.93 Fig. 3: BShap for Eq 13 demonstrated the potential of both IG and BShap in preserving risk patterns.","support":"partial"},{"finding":"Integrated Gradients cannot maintain demand monotonicity, i.e., attributions may not increase with a monotonic feature.","kind":"theoretical","quote":"IG, however, cannot maintain demand monotonicity, as if the model is monotonic in a certain feature, then the attributions for that feature increase as its value increases; Shapley value may involve calculations outside of the training domain, which could present difﬁculties such as times of ﬁnancial crisis.","support":"full"},{"finding":"Shapley value may involve calculations outside the training domain, posing difficulties during financial crises.","kind":"theoretical","quote":"IG, however, cannot maintain demand monotonicity, as if the model is monotonic in a certain feature, then the attributions for that feature increase as its value increases; Shapley value may involve calculations outside of the training domain, which could present difﬁculties such as times of ﬁnancial crisis.","support":"full"},{"finding":"BShap may not be the best option when its preservation properties are insufficient for a given situation.","kind":"negative","quote":"Therefore, BShap may not be the best option in this situation. VI.","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":18,"citations":{"cited_by_in_corpus":2,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":2},"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":"2026-11-14T19:56:30Z"},"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/"}