{"paper_id":"2411.05998","title":"Filling in Missing FX Implied Volatilities with Uncertainties: Improving VAE-Based Volatility Imputation","authors":["Achintya Gopal"],"paper_date":"2024-11-08","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2411.05998v1","page_url":"https://thequant.space/flowcharts/filling-in-missing-fx-implied-volatilities-with-uncertaintie/","scores":{"empirical_rigor":6.5,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes simple architectural modifications to VAEs, including σ‑VAE, Σ‑VAE and residual connections, to improve imputation of missing FX implied volatilities and provide accurate uncertainty estimates.","quote":"We show that simple modifications to the architecture of the VAE lead to significant imputation performance improvements (e.g., in low missingness regimes, nearly cutting the error by half), removing the necessity of using $β$-VAEs.","summary":"Missing data is a common problem in finance and often requires methods to fill in the gaps, or in other words, imputation. In this work, we focused on the imputation of missing implied volatilities for FX options. Prior work has used variational autoencoders (VAEs), a neural network-based approach…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Training data may contain missing values and must be handled by marginalizing out missing variables.","type":"data","stated":"explicit","quote":"Since the overall goal of our paper is to use VAEs for filling in missing data, we expect that our training data will also contain missing values."},{"assumption":"The volatility surface is represented as a discrete set of observations.","type":"data","stated":"explicit","quote":"For this work, since our surface has a discrete number of observations, we use:"},{"assumption":"Accurate posterior estimates require refitting the encoder on the specific dataset.","type":"model","stated":"explicit","quote":"To achieve an accurate estimate of the posterior, we can re-optimize (refit) the encoder q(z|xo) (Mattei and Frellsen, 2018b) using: argmin_ϕ LVAE(xo; σx, θ, ϕ)"},{"assumption":"No-arbitrage constraints are not enforced during training; model performance on arbitrage is evaluated post‑hoc.","type":"other","stated":"explicit","quote":"Note that we do not include any constraints during training to enforce no-arbitrage conditions, the performance we see is from what the model learned from the data."}]},"datasets":{"status":"available","items":[{"name":"FX volatility surfaces (AUDUSD, EURUSD, USDCAD, GBPCAD, USDMXN)","provider":"Bloomberg","kind":"options","access":"commercial","period":null,"universe":null,"quote":"σ-VAE w/o Residual 35.8 ± 0.1 24.5 ± 0.1 34.6 ± 0.1 21.6 ± 0.1 38.6 ± 0.1 σ-VAE w/ Residual 78.9 ± 0.1 73.7 ± 0.1 87.7 ± 0.1 65.9 ± 0.1 76.8 ± 0.1","known_id":"bloomberg"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"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":"Residual blocks halve the MAE for low missingness (10% missingness: 19.9 bps without residuals vs 8.4 bps with residuals).","kind":"performance","quote":"using residual blocks significantly outperforms all other models across missingness rates with the MAE dropping by more than half for 10% missingness rate (19.9 bps without residual blocks to 8.4 bps with residual blocks, Figure 14)."},{"finding":"σ‑VAE outperforms β‑VAE and classical Heston models, especially at higher missingness rates.","kind":"performance","quote":"similar to previous findings (Bergeron et al., 2022), the σ-VAE outperforms Heston with the largest gap in performance at higher rates of missingness."},{"finding":"Heston++ outperforms σ‑VAE for missingness rates below 50%, but residual‑net σ‑VAE dominates at higher missingness.","kind":"performance","quote":"Heston++ outperforms σ-VAE for missingness rates less than 50% (Figure 14), with the gap being much larger in certain tenors (Figure 15)."},{"finding":"Σ‑VAE achieves similar imputation error to σ‑VAE while providing better uncertainty estimates.","kind":"performance","quote":"the imputation performance of Σ-VAE is nearly identical to σ-VAE; while we might expect the performance to improve, the main intention of using Σ-VAE is to improve the uncertainty estimates."},{"finding":"Including residual connections and Σ‑VAE improves the proportion of generated surfaces that are arbitrage‑free.","kind":"performance","quote":"Table 2: Percentage of surfaces which contain no-arbitrage opportunities"}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":25,"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":"2026-11-14T17:55:24Z"},"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/"}