Pith. sign in

REVIEW 3 major objections 6 minor 12 references

A Sinusoidal Hull-White Model for Interest Rate Dynamics: Capturing Long-Term Periodicity in U.S. Treasury Yields

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A sinusoidal mean-reversion speed improves Hull-White pricing of 30-year Treasury bonds.

desk verdict A small, honestly reported model extension whose central empirical claim collapses under its own calibrated parameters and in-sample evaluation. read the letter →

arxiv 2506.06317 v1 pith:ZBEMAC6A submitted 2025-05-27 q-fin.ST q-fin.CP

Amit Kumar Jha This is my paper
classification q-fin.STq-fin.CP MSC 91G3091G6091B8460H1062P0591G2091G70
keywords interestratemodelingHull-WhitemodelperiodicmeanreversioncyclicalratesyieldcurvecalibrationtermstructureMonteCarlosimulationzero-couponbondpricing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes an extension of the Hull-White short-rate model in which the mean-reversion speed oscillates sinusoidally, as κ_t = κ0 + A sin(ωt), instead of remaining constant. The motivation is a Fourier analysis of daily U.S. Treasury yields from 1990 to 2022 that finds dominant cycles near 3.7, 5.5, 11, and 22 years, with the 22-year cycle most persistent at long maturities. On calibration, the sinusoidal model reports a root-mean-square yield error of 0.12% versus 0.14% for the standard model, and a 30-year zero-coupon bond price of 0.4864, closer to the observed 0.4926 than the standard model's 0.4862. The paper itself notes that the calibrated angular frequency came out at about 0.97 days rather than the intended 22-year period, and it recommends constraining ω in future work.

What carries the argument

The load-bearing object is the time-varying mean-reversion coefficient κ_t = κ0 + A sin(ωt), with ω intended to be 2π/(22×365) ≈ 0.00078 radians per day so that one full cycle lasts 22 years. Inside the Hull-White short-rate SDE, this coefficient replaces the constant κ, and bond prices lose their closed-form simplicity; the paper prices zero-coupon bonds through first-order Euler discretization with Monte Carlo simulation and calibrates κ0, A, ω, θ, and σ by a derivative-free simplex optimization routine against observed bond prices. What the mechanism does is allow the speed of reversion to the long-term mean to strengthen and weaken over time, which is how the model expresses the empirically detected long-period cycle.

What would settle it

Re-estimate the sinusoidal model with ω fixed at 2π/(22×365) and compare RMSE and the 30-year bond price; if the constrained model no longer beats the standard Hull-White RMSE of 0.14% or moves the 30-year price away from 0.4926, the claimed long-cycle benefit is not supported. A second direct check is to run the same calibration on a different 30-year Treasury window and see whether the 0.97-day ω, rather than the 22-year ω, is what drives the improvement.

Watch

Extended reading notes

Core claim

The central discovery claimed is that adding a sinusoidal time-dependent mean-reversion speed to the Hull-White model lets the short rate absorb the long-term cyclical behavior visible in U.S. Treasury yields. With parameters estimated by minimizing squared bond-price errors across eight maturities, the paper reports an RMSE of 0.12% versus 0.14% for the standard Hull-White model and a 30-year zero-coupon bond price of 0.4864 that is closer to the observed 0.4926 than the standard model's 0.4862. The paper presents this as evidence that periodic mean reversion improves the fit mainly at the long end, where the constant-speed model misprices the 30-year bond by a small but systematic amount.

Load-bearing premise

The load-bearing premise is that a mean-reversion speed oscillating with a period of about one day faithfully implements the 22-year cycle the paper identifies in the data; if that frequency mismatch is real, the improved RMSE could be a numerical artifact rather than evidence of long-term periodicity.

Editorial extensions

If this is right

  • Long-dated instruments such as 30-year bonds would be priced more accurately if the mean-reversion speed is allowed to vary cyclically rather than held constant.
  • The calibrated non-zero amplitude (A = 0.2110) would indicate that periodicity is not a negligible component of U.S. Treasury yield dynamics.
  • Interest-rate risk measures at long horizons would inherit a cyclical component, affecting duration and risk estimates for pension and insurance portfolios.
  • If the calibration is rerun with ω fixed to the 22-year frequency, the reported improvement may persist or vanish; the current reported gain is conditional on the unconstrained, near-daily ω.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A fair test of the 22-year-cycle story is to constrain ω to 2π/(22×365) and check whether the 30-year improvement survives without worsening the 20-year error, which the paper already reports as larger than the standard model's.
  • The 30-year bond-price difference between the two models is only 0.0002, well within Monte Carlo noise at 200 paths, so the economic significance of the long-end improvement likely needs many more simulation paths or variance reduction.
  • The sinusoidal coefficient could be transplanted into a short-rate framework that keeps rates positive, such as a CIR process, which would remove the negative-rate paths the paper displays and would test whether the periodicity benefit is robust to the choice of model.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes extending the Hull-White short-rate model by making the mean reversion speed sinusoidal, κ_t = κ0 + A sin(ωt), motivated by a Fourier analysis of daily U.S. Treasury yields (1990–2022) that identifies periodicities, including a 22-year cycle. The standard Hull-White affine bond price is derived, and the paper shows that the sinusoidal extension lacks a closed-form solution, so Monte Carlo simulation is used for pricing. Calibration is performed with Nelder-Mead on eight maturities, and the paper reports an RMSE improvement from 0.14% to 0.12% and a 30-year zero-coupon bond price of 0.4864 under the sinusoidal model versus 0.4862 under the standard model, concluding that the model captures long-term periodicity.

Significance. If the central empirical claim held, the model would offer a simple and tractable way to incorporate long-term cyclicality into a short-rate framework, with potential value for pricing long-dated instruments. The paper deserves credit for correctly deriving the standard Hull-White bond price, for transparently showing why the sinusoidal extension lacks a closed-form solution, and for reporting Monte Carlo validation errors. However, the load-bearing empirical evidence fails: the calibrated frequency is not the 22-year cycle, the claimed RMSE improvement is contradicted by the paper's own error tables, and the reported bond-price improvement is within Monte Carlo error. As a result, the paper does not demonstrate that periodic mean reversion captures long-term periodicity.

major comments (3)
  1. [Sections 5.3 and 6.3] The model's defining feature is ω = 2π/(22×365) ≈ 0.00078 rad/day, and Section 5.3 states that ω is fixed at this value. The reported calibration, however, yields ω = 6.4407 rad/day, corresponding to a period of about 0.97 days, which the paper itself acknowledges is inconsistent with the 22-year cycle. Consequently, the fitted model is not the long-periodicity model, and any improvement in fit cannot be attributed to long-term periodicity; it is a high-frequency artifact. The central claim of the paper is therefore unsupported. The authors must re-run the calibration with ω fixed at 0.00078 and report those results before the empirical claim can be evaluated.
  2. [Section 6.5, Tables 4 and 5] The abstract, introduction, and conclusion report RMSE values of 0.12% for the sinusoidal model and 0.14% for the standard model. However, squaring and averaging the errors in Table 5 (sinusoidal model) gives RMSE ≈ 0.143%, which is slightly larger than the standard model's ≈ 0.137% computed from Table 4. The claimed RMSE improvement is therefore not supported by the paper's own numerical tables; this is a load-bearing numerical inconsistency.
  3. [Sections 6.4 and 5.4] The 30-year bond price improvement is 0.4864 versus 0.4862, a difference of 0.0002, yet Section 5.4 states that Monte Carlo validation errors are typically maintained below 0.005. The difference between the two models is an order of magnitude smaller than the numerical error of the pricing method. Moreover, the RMSE is computed in-sample on the same eight maturities used for calibration, so the comparison provides no out-of-sample evidence of predictive improvement. The numerical support for the model is thus both statistically and methodologically weak.
minor comments (6)
  1. [Abstract] The abstract states that the 30-year zero-coupon bond price is 0.43 for the sinusoidal model and 0.47 for the standard model, but Section 6.4, Table 3 reports 0.4864 and 0.4862, respectively.
  2. [Section 5.3] The text both states that ω is fixed at 0.00078 and includes ω in the free parameter vector; the calibration description should be reconciled with the actual procedure that yields ω = 6.4407.
  3. [Section 4] The Fourier analysis is described as applied to detrended series, but the detrending method is not specified. Given that the ADF tests indicate non-stationarity, the spectral peaks may be artifacts; the authors should state the detrending procedure and its implications.
  4. [Section 5.4] Using 200 Monte Carlo paths and a time step of 0.05 years is quite coarse for 30-year bonds; the paper should discuss the standard error of the simulated bond prices, especially because the claimed improvement is only 0.0002.
  5. [References] The Duffie and Kan entry is dated 2000, but the DOI (10.1111/j.1467-9965.1996.tb00123.x) corresponds to 1996; please correct the year.
  6. [Throughout] The paper alternates between first-person singular ('I') and plural ('we'), for example in Section 4; please standardize the voice.

Circularity Check

2 steps flagged · score 6.0 of 10

Reported fit improvement is in-sample and uses a freely calibrated ω=6.4407 rad/day that contradicts the 22-year premise; the central performance claim reduces to the calibration objective.

  1. fitted input called prediction [Section 5.3 (Calibration Procedure) and Section 6.4 (Bond Price Comparison), Table 3]
    "“The calibration process minimizes the sum of squared differences between the model-implied bond prices and the observed bond prices derived from the yield curve. ... The optimization problem is formally defined as: min θ Σ_{Ti} (Pmodel(0, Ti; θ) − Pobserved(Ti))^2 ... The sinusoidal HW model shows a closer fit to the observed price at 30 years (0.4864 vs. 0.4926), compared to the standard HW model (0.4862).”"

    The same eight maturities (1, 2, 3, 5, 7, 10, 20, 30 years) appear both in the Nelder-Mead objective and in the reported RMSE/30-year comparison. The sinusoidal model's price at 30 years is one of the fitted points, so “0.4864 vs. 0.4926” and the 0.12%-vs-0.14% RMSE are in-sample properties of the optimization, not out-of-sample predictions. Presenting the fitted bond price as evidence that periodic mean reversion “captures long-term yield dynamics” restates the minimization target rather than independently confirming the model.

  2. other [Section 5.2 (Sinusoidal HW Model), Section 5.3 (Calibration Procedure), Section 6.3 (Calibration Results)]
    "“with ω = 2π/(22×365) ≈ 0.00078 radians/day, tuned to the 22-year cycle from Section 4 ... the calibrated parameters are κ0 = 0.3068, A = 0.2110, ω = 6.4407 ... corresponding to a period of approximately 0.97 days, appears inconsistent with the 22-year periodicity ... This discrepancy may indicate a calibration artifact ... suggests that ω should be constrained to 0.00078 in future iterations.”"

    The independent Fourier evidence (22-year cycle) is invoked as the reason for the sinusoidal ansatz, but ω is not actually held at the empirically determined value. The calibrated model that produces the reported prices uses ω=6.4407 rad/day, a ~0.97-day period, so the fit improvement cannot be attributed to the hypothesized long-term cycle. The paper itself labels the result a “calibration artifact” and says future work should fix ω, confirming that the current numerical support for the central 22-year-periodicity claim is not provided by the calibrated model.

full rationale

The derivation chain is: (i) Fourier analysis finds a 22-year cycle; (ii) the paper proposes κ_t = κ0 + A sin(ωt) with ω nominally set to 0.00078; (iii) Nelder-Mead calibrates all parameters by minimizing squared bond-price errors on the observed yield curve; (iv) the resulting in-sample RMSE and 30-year price are reported as evidence that the sinusoidal model captures long-term periodicity. Step (iv) is not independent of step (iii): the “improved” prices are the optimized values of the same objective, and the calibrated ω (6.4407) is the one the paper says it intended to avoid. Thus the central empirical claim reduces, by the paper's own equations, to a free fit plus an internal inconsistency rather than to the independent periodicity evidence or to a constrained-ω test. The external references (Bauer and Rudebusch, Campbell et al., etc.) are used only as background motivation and are not load-bearing circular citations. There is no self-citation chain or imported uniqueness theorem. The corrections needed (fix ω at 0.00078, add out-of-sample validation) are correctly identified by the paper itself, but they are future work, not current support. Score 6: one or more reported “predictions”/fit comparisons reduce by construction to the calibration objective, and the fitted frequency contradicts the stated 22-year premise.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The model rests on five fitted parameters and several domain assumptions. The most consequential are the Gaussian short-rate framework, the spectral interpretation of Fourier peaks as stable cycles, and the accuracy of a 200-path Monte Carlo estimator. No new entities are introduced.

free parameters (5)
  • κ0 = 0.3068
    Baseline mean-reversion speed in the sinusoidal Hull-White model, estimated by Nelder-Mead on eight FRED yields from December 2022.
  • A = 0.2110
    Amplitude of the sinusoidal mean-reversion term, estimated; this parameter controls how strongly the pull varies over time.
  • ω = 6.4407 rad/day
    Angular frequency of the sinusoidal term. Intended to be fixed at 2π/(22×365)≈0.00078, but the optimizer returned 6.4407, corresponding to a period of about 0.97 days. This fitted value is what generated the reported bond prices.
  • θ = 0.0256
    Long-term mean short rate, estimated in the calibration.
  • σ = 0.0101
    Short-rate volatility, estimated in the calibration.
assumptions (5)
  • domain assumption The short rate follows a Gaussian diffusion dr_t = κ(θ−r_t)dt + σ dW_t under the risk-neutral measure.
    Invoked in Section 5.1. This assumption permits negative short rates, which the paper later acknowledges as a limitation.
  • standard math Zero-coupon bond prices satisfy the Feynman-Kac PDE with terminal condition P(T,T)=1.
    Used in Sections 5.1 and 5.2 to derive the bond pricing equations from the short-rate SDE.
  • domain assumption Dominant peaks in the Fourier magnitude spectrum of detrended yields correspond to stable, deterministic periodic components suitable for a sinusoidal mean-reversion parameter.
    Section 4 uses FT peaks (3.7, 5.5, 11, 22 years) to motivate κ_t = κ0 + A sin(ωt), but no statistical test of stability or significance is given.
  • domain assumption Euler-Maruyama with Δt=0.05 and 200 paths yields bond-price estimates accurate enough to compare models.
    Section 5.4. The paper's own validation shows MC errors up to 0.005, which are larger than the reported model differences.
  • domain assumption Observed zero-coupon bond prices can be bootstrapped as exp(−T·y_T) from par yields.
    Section 5.3. This ignores coupon effects and uses a single yield-curve snapshot for calibration.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Sinusoidal Hull-White Model for Interest Rate Dynamics: Capturing Long-Term Periodicity in U.S. Treasury Yields." pith.science (2026). https://pith.science/paper/ZBEMAC6A

@misc{pith2026250606317,
  author       = {Pith},
  title        = {Pith review of: A Sinusoidal Hull-White Model for Interest Rate Dynamics: Capturing Long-Term Periodicity in U.S. Treasury Yields},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZBEMAC6A}},
  note         = {Machine review of arXiv:2506.06317}
}
read the original abstract

This study is motivated by empirical observations of periodic fluctuations in interest rates, notably long-term economic cycles spanning decades, which the conventional Hull-White short-rate model fails to adequately capture. To address this limitation, we propose an extension that incorporates a sinusoidal, time-varying mean reversion speed, allowing the model to reflect cyclic interest rate dynamics more effectively. The model is calibrated using a comprehensive dataset of daily U.S. Treasury yield curves obtained from the Federal Reserve Economic Data (FRED) database, covering the period from January 1990 to December 2022. The dataset includes tenors of 1, 2, 3, 5, 7, 10, 20, and 30 years, with the most recent yields ranging from 1.22% (1-year) to 2.36% (30-year). Calibration is performed using the Nelder-Mead optimization algorithm, and Monte Carlo simulations with 200 paths and a time step of 0.05 years. The resulting 30-year zero-coupon bond price under the proposed model is 0.43, compared to 0.47 under the standard Hull-White model. This corresponds to root mean squared errors of 0.12% and 0.14%, respectively, indicating a noticeable improvement in fit, particularly for longer maturities. These results highlight the model's enhanced capability to capture long-term yield dynamics and suggest significant implications for bond pricing, interest rate risk management, and the valuation of interest rate derivatives. The findings also open avenues for further research into stochastic periodicity and alternative interest rate modeling frameworks.

Figures

Figures reproduced from arXiv: 2506.06317 by the authors.

Figure 1
Figure 1. Time series and ACF plots for U.S. Treasury yields across maturities (1, 2, 3, 5, 7, 10, [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Fourier Transform magnitude spectra for U.S. Treasury yields across maturities, high [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Comparison of observed and fitted yields for the standard Hull-White (blue dashed line) [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Bond price comparison (top left), error analysis (top right), and simulated short-rate [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

12 extracted references · 6 canonical work pages

  1. [6]

    Federal Reserve Economic Data

    doi: 10.1111/j.1467-9965.1996.tb00123.x. Federal Reserve Economic Data. U.s. treasury yields.https://fred.stlouisfed.org/,

    arXiv 1996

  2. [1985]

    Qiang Dai and Kenneth J

    doi: 10.2307/1911242. Qiang Dai and Kenneth J. Singleton. Specification analysis of affine term structure models.Journal of Finance, 55(5):1943–1978,

    doi:10.2307/1911242 1943

  3. [1989]

    19 David Heath, Robert Jarrow, and Andrew Morton

    doi: 10.2307/1912559. 19 David Heath, Robert Jarrow, and Andrew Morton. Bond pricing and the term structure of interest rates: A new methodology for contingent claims valuation.Econometrica, 60(1):77–105,

    doi:10.2307/1912559

  4. [1990]

    John Hull and Alan White

    doi: 10.1093/rfs/3.4.573. John Hull and Alan White. Numerical procedures for implementing term structure models ii: Two-factor models.Journal of Derivatives, 2(2):37–48,

    work page doi:10.1093/rfs/3.4.573

  5. [1991]

    Mike West and Jeff Harrison.Bayesian Forecasting and Dynamic Models

    doi: 10.3905/jfi.1991.692347. Mike West and Jeff Harrison.Bayesian Forecasting and Dynamic Models. Springer, New York, NY, 2 edition,

    work page arXiv 1991

  6. [1992]

    John Hull.Options, Futures, and Other Derivatives

    doi: 10.2307/2951677. John Hull.Options, Futures, and Other Derivatives. Prentice Hall, Upper Saddle River, NJ, 6 edition,

    doi:10.2307/2951677

  7. [1994]

    Noureddine Krichene

    doi: 10.3905/jod.1994.407912. Noureddine Krichene. Recent dynamics of g7 bond yields: A spectral perspective. Working Paper WP/06/208, International Monetary Fund,

    work page arXiv 1994

  8. [1996]

    doi: 10.1017/S0266466600006976. James D. Hamilton. A new approach to the economic analysis of nonstationary time series and the business cycle.Econometrica, 57(2):357–384,

    doi:10.1017/s0266466600006976

Show all 12 references
  1. [2000]

    Francis X

    doi: 10.1111/0022-1082.00278. Francis X. Diebold and Canlin Li. Forecasting the term structure of government bond yields. Journal of Econometrics, 130(2):337–364,

  2. [2006]

    Darrell Duffie and Rui Kan

    doi: 10.1016/j.jeconom.2005.03.005. Darrell Duffie and Rui Kan. A yield-factor model of interest rates.Mathematical Finance, 6(4): 379–406,

    2005 doi

  3. [2008]

    Michael D

    doi: 10.1198/073500108000000068. Michael D. Bauer and Glenn D. Rudebusch. Interest rates under falling stars.American Economic Review, 108(10):3123–3154,

    doi

  4. [2018]

    Damiano Brigo and Fabio Mercurio.Interest Rate Models - Theory and Practice: With Smile, Inflation and Credit

    doi: 10.1257/aer.20170625. Damiano Brigo and Fabio Mercurio.Interest Rate Models - Theory and Practice: With Smile, Inflation and Credit. Springer, Berlin, Heidelberg, 2 edition,

    doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.