{"paper_id":"2509.11928","title":"Meta-Learning Neural Process for Implied Volatility Surfaces with SABR-induced Priors","authors":["Jirong Zhuang","Xuan Wu"],"paper_date":"2025-09-15","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2509.11928v3","page_url":"https://thequant.space/flowcharts/meta-learning-neural-process-for-implied-volatility-surfaces/","scores":{"empirical_rigor":8,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes the Volatility Neural Process (VolNP), an attention‑based neural process that is pre‑trained on SABR‑generated data and fine‑tuned on market data to map sparse option quotes to a full implied volatility surface in a single forward pass.","quote":"we propose the Volatility Neural Process (VolNP), a neural process model (Garnelo et al., 2018a,b) that maps a small context set of option quotes to a full surface via conditional prediction.","summary":"We treat implied volatility surface (IVS) reconstruction as a learning problem guided by two principles. First, we adopt a meta-learning view that trains across trading days to learn a procedure that maps sparse option quotes to a full IVS via conditional prediction, avoiding per-day calibration at…","strategy":"Volatility Neural Process (VolNP)"},"assumptions":{"status":"available","items":[{"assumption":"Model parameters are fixed at test time with no per‑day updates, avoiding daily recalibration.","type":"model","stated":"explicit","quote":"At test time, parameters are fixed (i.e., with no per-day updates), hence daily recalibration is avoided."},{"assumption":"The synthetic SABR surface provides a geometric prior that is realistic for market smiles.","type":"model","stated":"explicit","quote":"The SABR model is calibrated to each maturity slice in Dq, and parameters are linearly interpolated across the term structure."},{"assumption":"Option quotes are filtered for liquidity and quality (positive bid/ask, bid<ask, volume>5, spread≤50%, price≥0.1).","type":"data","stated":"explicit","quote":"Quotes are preprocessed by computing mid-prices from best bid/ask and applying liquidity filters: positive bid and ask with bid<ask; daily volume > 5; relative bid-ask spread ≤50%; and mid-price ≥0.1."},{"assumption":"No‑butterfly‑arbitrage condition must hold for the implied risk‑neutral density.","type":"model","stated":"explicit","quote":"We test for violations of the no-butterfly-arbitrage condition, which ensures the model’s implied risk-neutral probability density is non-negative."}]},"datasets":{"status":"available","items":[{"name":"SPX options","provider":"OptionMetrics","kind":"options","access":"commercial","period":"January 2006 to August 2023","universe":"S&P 500 index options","quote":"Our experiments are carried out on a dataset of S&P 500 (SPX) index options, spanning from January 2006 to August 2023. The raw data is from OptionMetrics dataset...","known_id":"optionmetrics"},{"name":"Synthetic SABR dataset","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"The second is a synthetic dataset, { ˆDq}q∈Q, where ˆDq is the corresponding dense, smooth volatility surface generated by a calibrated SABR model."}],"known":[{"id":"optionmetrics","name":"OptionMetrics IvyDB","provider":"OptionMetrics","kind":"options","access":"commercial","quote":"Forward prices are computed from the zero-coupon curve and dividend yields (from OptionMetrics), and implied volatilities are obtained under the Black framework.","mentions":2,"used_in":2}],"sample":{"start":"2006","end":"2023","market":"S&P 500 index options","asset_class":"options","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":"pricing-calibration","findings":[{"finding":"VolNP‑SP achieves the lowest overall RMSE (96.04 BPS) and MAE (48.22 BPS) with R² = 0.98, outperforming all benchmarks.","kind":"performance","quote":"Overall, VolNP‑SP achieves the best performance, with the lowest RMSE (96.04 BPS) and MAE (48.22 BPS), and the highest R2 (0.98)."},{"finding":"Pre‑training on SABR reduces RMSE by nearly 40% compared to the market‑only baseline (VolNP‑Base).","kind":"predictive-accuracy","quote":"While their overall MAE is similar, the pre‑training strategy allows VolNP‑SP to reduce the RMSE by nearly 40%."},{"finding":"VolNP‑SP shows superior performance for mid‑ and long‑term maturities where data are sparse, while GP performs best only for short‑term dense data.","kind":"performance","quote":"In these mid- and long‑term regimes, VolNP‑SP outperforms all other models. This indicates that the learned geometric prior is crucial for constructing a coherent and robust surface, especially in regions with fewer data points."},{"finding":"VolNP‑SP surfaces are free of no‑butterfly‑arbitrage violations, unlike the ablation model VolNP‑Base.","kind":"other","quote":"Smiles from VolNP‑Base (bottom row) exhibit significant violations of this condition... In"},{"finding":"VolNP‑SP is computationally efficient, requiring only a single forward pass per day (≈ 0.02 s on a RTX 3060).","kind":"performance","quote":"VolNP models run in PyTorch (CUDA) and we time a single forward pass"}],"reported_performance":{"strategy":"Volatility Neural Process (VolNP)","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":48,"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":1,"extracted_at":"2026-11-14T17:27:11Z"},"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/"}