{"paper_id":"2511.17892","title":"Arbitrage-Free Bond and Yield Curve Forecasting with Neural Filters under HJM Constraints","authors":["Xiang Gao","Cody Hyndman"],"paper_date":"2025-11-22","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2511.17892v1","page_url":"https://thequant.space/flowcharts/arbitrage-free-bond-and-yield-curve-forecasting-with-neural/","scores":{"empirical_rigor":7.5,"math_complexity":9.2,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes an arbitrage‑free deep learning framework for yield‑curve and bond‑price forecasting that embeds HJM no‑arbitrage drift restrictions into a neural state‑space architecture with Kalman, extended Kalman and particle filters and an explicit arbitrage error regularization term.","quote":"We develop an arbitrage-free deep learning framework for yield curve and bond price forecasting based on the Heath-Jarrow-Morton (HJM) term-structure model and a dynamic Nelson-Siegel parameterization of forward rates.","summary":"We develop an arbitrage-free deep learning framework for yield curve and bond price forecasting based on the Heath-Jarrow-Morton (HJM) term-structure model and a dynamic Nelson-Siegel parameterization of forward rates. Our approach embeds a no-arbitrage drift restriction into a neural state-space…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The forward‑rate model obeys a no‑arbitrage drift restriction derived from the HJM framework.","type":"model","stated":"explicit","quote":"Our approach embeds a no-arbitrage drift restriction into a neural state-space architecture by combining Kalman, extended Kalman, and particle ﬁlters with recurrent neural networks (LSTM/CLSTM)."},{"assumption":"The state variable Xt follows a mean‑reverting stochastic differential equation.","type":"model","stated":"explicit","quote":"We deﬁne Xt as extended Vašíˇcek process dXt = κt (θt −Xt) dt + σtdWt"},{"assumption":"Only fixed‑coupon, non‑callable, non‑convertible bonds with remaining maturity between 3 months and 30 years are used.","type":"data","stated":"explicit","quote":"We retain ﬁxed-coupon, non-callable, non-convertible bonds with remaining time-to-maturity between 3 months and 30 years, yield-to-maturity under 700 bps, and an absolute YTM–coupon difference under 500 bps."},{"assumption":"The decay parameter λ is selected by grid‑search minimizing out‑of‑sample RMSE and then kept fixed across the sample.","type":"statistical","stated":"explicit","quote":"We select the decay parameter λ by a grid search minimizing out-of-sample RMSE over a rolling validation window and keep it ﬁxed across the sample for stability."},{"assumption":"Observation noise for price forecasting follows a multivariate generalized Gaussian distribution (MGGD).","type":"model","stated":"explicit","quote":"Because price errors exhibit heavy tails and occasional outliers, we model observation noise with a multivariate generalized Gaussian distribution (MGGD), which nests Gaussian…"}]},"datasets":{"status":"available","items":[{"name":"FINRA‑TRACE and proprietary Treasury data feed","provider":"FINRA‑TRACE (supplemented by commercial vendor)","kind":"market-prices","access":"commercial","period":"2017–2019","universe":"U.S. Treasuries and corporate bonds","quote":"Raw data were obtained from FINRA-TRACE supplemented by proprietary Treasury data feed from a commercial vendor. Our data comprise daily clean prices, yields to maturity, coupon rates/frequencies, instrument type, convertibility/callability flags, and issue/maturity dates for U.S. Treasuries and corporate bonds from 20"}],"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":"forecast-accuracy","findings":[{"finding":"Arbitrage regularization yields the largest predictive gains at short maturities and the 5‑day horizon, improving bid‑ask hit rates while keeping MAE/RMSE competitive.","kind":"performance","quote":"Empirically, AER delivers the largest gains at short maturities and at the 5-day horizon, improving bid–ask hit rates while keeping MAE/RMSE competitive (Section 5)."},{"finding":"The arbitrage‑regularized model shows strongest improvements for short‑maturity forecasts, especially 5‑day‑ahead, reducing dollar‑denominated prediction errors.","kind":"predictive-accuracy","quote":"Arbitrage regularization leads to its strongest improvements at short maturities, particularly in 5-day-ahead forecasts, increasing market‑consistency as measured by bid‑ask hit rates and reducing dollar‑denominated prediction errors."},{"finding":"Particle filtering with MGGD observation noise achieves high effective sample size and stable forecasting performance despite heavy‑tailed price errors.","kind":"performance","quote":"We conclude that the particle ﬁlter with multivariate generalized Gaussian distribution is very efﬁcient and stable for bond prices forecasting in both training and testing data."},{"finding":"For corporate bond data, the Kalman filter outperforms EKF and PF models, but overall errors are larger than for Treasury data due to limited sample and omitted credit risk factors.","kind":"predictive-accuracy","quote":"In the case of corporates the yield data contains seems to contain more information than the corporate bond prices and we ﬁnd that the KF model signiﬁcantly outperforms the EKF and PF models."},{"finding":"Different filters trade off accuracy and computation: KF in yield‑space is fastest, EKF offers a good accuracy‑runtime balance, PF is most demanding but handles non‑Gaussian residuals.","kind":"other","quote":"For practical use, different ﬁlters trade accuracy for computation. A KF in yield-space is fastest but requires extracting yields from prices. An EKF in price-space offers a good accuracy–runtime balance. A PF in price-space with MGGD observation noise and systematic resampling is the most computationally demanding but"}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":33,"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":"2026-11-14T17:21:06Z"},"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/"}