{"paper_id":"2601.04608","title":"Forecasting the U.S. Treasury Yield Curve: A Distributionally Robust Machine Learning Approach","authors":["Jinjun Liu","Ming-Yen Cheng"],"paper_date":"2026-01-08","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2601.04608v1","page_url":"https://thequant.space/flowcharts/forecasting-the-u.s.-treasury-yield-curve-a-distributionall/","scores":{"empirical_rigor":7,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a distributionally robust ensemble forecasting framework that combines a rolling‑window Factor‑Augmented Dynamic Nelson‑Siegel model with high‑dimensional Random Forest forecasts via adaptive forecast combinations.","quote":"We propose a distributionally robust ensemble forecasting framework that integrates parametric factor models with high dimensional nonparametric machine learning models through adaptive forecast combinations.","summary":"We study U.S. Treasury yield curve forecasting under distributional uncertainty and recast forecasting as an operations research and managerial decision problem. Rather than minimizing average forecast error, the forecaster selects a decision rule that minimizes worst case expected loss over an…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Only lagged predictor information is used to reflect data availability.","type":"data","stated":"explicit","quote":"Considering the release of many key economic data is lagging 1 month, only lagged predictor information is used to reflect data availability."},{"assumption":"Forecast errors are evaluated under a moment‑based ambiguity set allowing uncertainty in mean and covariance.","type":"model","stated":"explicit","quote":"DRO seeks decisions that perform well against the worst-case distribution within a prescribed ambiguity set."},{"assumption":"Ridge‑regularized covariance matrices stabilize second‑moment estimation in the combination step.","type":"statistical","stated":"explicit","quote":"instability in second‑moment estimation is mitigated through ridge‑regularized covariance matrices."},{"assumption":"The severity of the worst‑case criterion is adjustable, allowing the forecaster to regulate robustness versus efficiency.","type":"model","stated":"explicit","quote":"The severity of the worst case criterion is adjustable, allowing the forecaster to regulate the trade off between robustness and statistical efficiency."},{"assumption":"Structural breaks are present in the yield series, as indicated by rejected CUSUM tests.","type":"statistical","stated":"explicit","quote":"The null is strongly rejected across all maturities (p < 0.001), indicating the presence of structural"}]},"datasets":{"status":"available","items":[{"name":"U.S. Treasury zero‑coupon equivalent yields","provider":"LSEG Reuters Workspace","kind":"market-prices","access":"commercial","period":"January 2006 to August 2025","universe":"15 maturities from 3 months to 30 years","quote":"We use zero‑coupon equivalent yields on U.S. Treasury securities obtained from the LSEG Reuters Workspace. The sample spans January 2006 to August 2025 at a monthly frequency and covers 15 maturities ranging from 3 months to 30 years.","known_id":"refinitiv"},{"name":"High‑dimensional panel of economic indicators","provider":null,"kind":"macro","access":"unknown","period":null,"universe":"111 indicator series","quote":"Let Zt ∈Rp denote the vector of economic indicators observed at month t, where p = 111."},{"name":"Global sovereign benchmark bond yields (10‑year)","provider":"LSEG Reuters Workspace","kind":"market-prices","access":"commercial","period":"January 2010 onward","universe":"major economies including Canada, China, Germany, Japan, Malaysia, United Kingdom, United ","quote":"We extend the Random Forest forecasting framework to a cross‑country setting by examining 10‑year benchmark government bond yields for a set of major economies, including Canada, China, Germany, Japan, Malaysia, the United Kingdom, and the United States.","known_id":"refinitiv"}],"known":[{"id":"refinitiv","name":"Refinitiv / LSEG (Datastream, Eikon, Tick History)","provider":"LSEG","kind":"market-prices","access":"commercial","quote":"2 Methodology 2.1 Data 2.1.1 U.S. Treasury Yields We use zero-coupon equivalent yields on U.S. Treasury securities obtained from the LSEG Reuters Workspace.","mentions":2,"used_in":2}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"weekly-monthly"},"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":"Adaptive forecast combinations outperform individual models at short horizons, while Random Forest forecasts dominate at medium and longer horizons.","kind":"performance","quote":"Empirical results based on monthly data and forecast horizons from one to twelve months show that adaptive combinations outperform individual models at short horizons, whereas RF forecasts dominate at medium and longer horizons."},{"finding":"Random Forest models deliver stable RMSFE across horizons, with short‑end yields showing the highest errors (≈24‑25 bps) and long‑dated maturities the lowest (≈13‑14 bps).","kind":"predictive-accuracy","quote":"The RF model delivers stable forecast accuracy across horizons, with RMSFE remaining broadly flat from one‑month to twelve‑month horizons for all maturities"},{"finding":"The Factor‑Augmented Dynamic Nelson‑Siegel (FADNS) model improves short‑horizon performance relative to the baseline DNS model, reducing RMSFE by roughly 5‑15 bps at the one‑month horizon.","kind":"predictive-accuracy","quote":"... consistently reduce RMSFE relative to DNS across maturities. The gains are most pronounced at short and intermediate maturities, where RMSFE declines by roughly 5–15 bps compared to the baseline DNS model."},{"finding":"Both DNS and FADNS models exhibit rapidly increasing RMSFE for horizons beyond six months, performing worse than Random Forest models.","kind":"negative","quote":"While FADNS continues to outperform DNS at the three‑month horizon, forecast errors increase rapidly beyond six months for all PCA specifications. At the nine‑ and twelve‑month horizons, RMSFE exceeds 150 bps at the short end of the curve and remains above 120 bps even at long maturities. The results are much worse tha"}],"reported_performance":null,"out_of_sample":{"method":"walk-forward","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":53,"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":"2026-11-14T17:16:55Z"},"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/"}