Pith. sign in

REVIEW 3 major objections 5 minor 27 references

Without prior knowledge of structural change, average rolling- and expanding-window demographic forecasts for a robust default.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 11:02 UTC pith:UWF7DWOU

load-bearing objection Solid multi-country empirical guidance on a routine choice; the equal-weight hedge is practical but never formally tested against the better single scheme. the 3 major comments →

arxiv 2607.10527 v1 pith:UWF7DWOU submitted 2026-07-12 stat.AP

Age-specific demographic modeling and forecasting: Rolling window, expanding window, or both?

Sizhe Chen , Han Lin Shang This is my paper
classification stat.AP
keywords demographic forecastingexpanding windowrolling windowforecast accuracymortalityfertilityforecast combinationfunctional time series
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

When forecasting age-specific mortality and fertility, researchers routinely choose either a rolling window (fixed recent history) or an expanding window (all past data). This paper shows that neither choice wins uniformly across countries, sexes, and horizons: expanding often suits smoother female mortality and fertility series, while rolling more often suits male mortality that is more sensitive to recent shocks. The authors therefore recommend a tuning-free equal-weight average of the two schemes. Across 23 mortality and 16 fertility countries, the combination is rarely the worst, frequently matches or beats the better single scheme, and damps late-horizon volatility in both point and interval accuracy. The practical claim is simple: if you lack firm knowledge of structural stability, average both windows rather than gamble on one.

Core claim

In multi-country comparisons of age-specific mortality and fertility forecasts, the superior performance of either the rolling or the expanding window often persists across horizons, yet neither scheme dominates uniformly by sex, component, or country. An equal-weight combination of the two forecasts therefore provides a robust, tuning-free alternative that is never systematically inferior and frequently tracks or exceeds the better single scheme while reducing end-of-horizon volatility.

What carries the argument

Equal-weight ensemble (Eq. 3): the simple average of the rolling-window and expanding-window forecasts produced by the same functional time-series model, with no extra tuning parameter.

Load-bearing premise

A fixed half-and-half average is assumed to be a good enough hedge even though which window wins can change with horizon, sex, and the underlying model, and the paper does not test whether the average is statistically better than the better single scheme.

What would settle it

Re-run the same multi-country holdout design with formal horizon-wise tests of equal predictive ability; if the equal-weight combination is significantly worse than the better single scheme on a large share of country-horizon cells, the robustness claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper compares rolling-window and expanding-window estimation schemes for age-specific mortality and fertility forecasting within a common functional time-series framework (Hyndman–Ullah, K=6), and proposes a fixed equal-weight average of the two forecasts (Eq. 3). Using multi-country data from the Human Mortality Database (23 countries) and Human Fertility Database (16 countries), with a fixed 20-year hold-out, it evaluates point accuracy (RMSFE, MAFE) and interval accuracy (ECP/CPD, interval score) over horizons h=1…20. Horizon-wise win-count heatmaps and country panels (Australia/Canada mortality; Canada/Japan fertility) show that no single scheme dominates uniformly: expanding (or combined) often leads for female mortality and for fertility, while rolling (or combined) often leads for male mortality and for mortality intervals. An appendix repeats the exercise under Lee–Carter and APC benchmarks. The central recommendation is that, absent prior information on structural stability, the equal-weight combination is a robust, tuning-free default.

Significance. The question of rolling versus expanding windows is practically important and under-studied in demographic forecasting; a transparent multi-country design with both point and interval metrics, sex-specific mortality, and an LC/APC sensitivity appendix is a genuine contribution. The equal-weight combination is simple, reproducible, and operationally attractive as a hedge when leadership changes with horizon, sex, or component. Code availability further strengthens the work. The main limitation is that the robustness claim rests on descriptive win counts and visual tracking rather than formal tests of equal predictive ability, so the statistical status of the combination as a default remains incompletely established.

major comments (3)
  1. Abstract and §6 claim that the equal-weight combination (Eq. 3) is a robust default that is “never systematically inferior” and frequently matches or exceeds the better single scheme. The supporting evidence is horizon-wise win-count heatmaps (Figs. 4, 7, 10, 13) and country line plots. These do not test equal predictive ability (e.g., Diebold–Mariano, MCS, or SPA). The paper itself notes that leadership changes with horizon, sex, and model class, yet never tests whether Combined is statistically better than, or not worse than, the better of Rolling/Expanding. Formal horizon-wise tests (or at least paired error differences with uncertainty) are needed to underwrite the central claim.
  2. The LC/APC appendix (Figs. 16–27 and interval tables) shows Combined rarely wins and often sits between the two single schemes, while Rolling dominates under LC and APC for mortality and LC-ASFR. The abstract/§6 recommendation that Combined is a robust default is therefore model-class dependent. The main text should qualify the claim by model class and state when the combination is recommended versus when a single scheme is preferred under lower-dimensional benchmarks.
  3. Eq. (3) fixes the combination weight at 1/2 with no data-driven alternative or sensitivity. Given that leadership changes with h, sex, and component, a short check of alternative fixed weights (e.g., 0.3/0.7) or a simple validation-based weight would show whether equal weight is special or merely convenient. Without this, the “tuning-free” virtue is clear but the optimality of 1/2 is unexamined.
minor comments (5)
  1. §5.1: interval evaluation uses h=1…19 because of the SD validation requirement, while point evaluation uses h=1…20; state this consistently in figure captions and text to avoid confusion.
  2. Country selection (continuous series through 1950) is reasonable but should be justified more explicitly as a design choice that may favor longer-history methods; a brief note on excluded countries would help.
  3. Notation: Yt(ui) is used for both raw and log rates in places; clarify the scale on which MAFE/RMSFE and interval scores are computed (original vs log).
  4. Figures 11–12 and 14–15 are dense; ensuring consistent axis scales and a single legend placement would improve readability.
  5. A few typos and awkward phrases remain (e.g., “panel countries,” “mean change”); a light copy-edit pass would help.

Circularity Check

0 steps flagged

No circularity: purely empirical multi-country comparison of rolling, expanding, and equal-weight combined forecasts on held-out data; no derivation that reduces to a fitted quantity by construction.

full rationale

The paper proposes a fixed equal-weight average of rolling- and expanding-window forecasts (Eq. 3) and evaluates it against the two single schemes using standard point (RMSFE, MAFE) and interval (ECP/CPD, interval score) metrics on multi-country holdout sets (last 20 years) for mortality and fertility, plus an LC/APC sensitivity appendix. All claims are comparative empirical statements about which scheme wins more often across horizons and countries; there is no first-principles derivation, uniqueness theorem, or fitted parameter that is then re-labeled as a prediction. Self-citations (Hyndman & Shang, Shang & Haberman, Shang & Xu) supply the functional time-series method, smoothing, and interval construction tools; they are not load-bearing for the target claim that the ½ combination is a robust default. The equal weight is an explicit, untuned ansatz, not a quantity fitted to the evaluation data. No circular steps exist.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

The paper rests on standard functional data analysis and classical demographic models; free parameters are conventional choices (K=6, equal weights, 20-year hold-out) rather than quantities fitted to produce the central claim. No new physical or statistical entities are postulated.

free parameters (4)
  • number of functional principal components K = 6
    Fixed at K=6 following Hyndman et al. (2013); affects the approximation quality of the age surface.
  • equal combination weight = 0.5
    Hard-coded ½ in Eq. (3); not optimised or cross-validated.
  • hold-out length n_test = 20
    Last 20 calendar years reserved for evaluation for every country.
  • interval nominal level α = 0.2 / 0.05
    Customarily 0.2 or 0.05 for interval scores; validation segment used to tune SD adjustment ξ.
axioms (3)
  • domain assumption Functional principal-component decomposition plus univariate ETS forecasts of scores yield valid multi-step age-specific forecasts (Hyndman–Ullah framework).
    Invoked throughout §3 as the common forecasting engine for all three schemes.
  • domain assumption Smoothing with monotonicity (mortality) or concavity (fertility) constraints improves subsequent forecast accuracy.
    Stated in §2 and used for all main results; raw-data variants are mentioned but not fully reported.
  • ad hoc to paper Countries with continuous national series through 1950 form a representative comparison set.
    Selection rule in §2.1–2.2; Belgium and others excluded for gaps.

pith-pipeline@v1.1.0-grok45 · 25398 in / 2423 out tokens · 29272 ms · 2026-07-14T11:02:48.873931+00:00 · methodology

0 comments
read the original abstract

Rolling and expanding windows are widely used in age-specific demographic modeling and forecasting. Building on these approaches, we propose a simple combination method that assigns equal weight to the forecasts from both schemes. Our focus is on evaluating and comparing the forecast accuracy of the two window types in modeling age-specific mortality and fertility. Based on the multi-country comparison, the superior performance of one method often persists across different forecast horizons. In the absence of prior information, our combined approach offers a robust and practical alternative.

Figures

Figures reproduced from arXiv: 2607.10527 by Han Lin Shang, Sizhe Chen.

Figure 1
Figure 1. Figure 1: Rainbow Plots: Australian age-specific female and male mortality rates from 1921 to 2021 with and without smoothing by single years of age from 0 to 94 and the last age group of 95+. The data from the distant past are shown in red, while the data in the most recent years are shown in purple. 4 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Rainbow Plots: Canadian female and male mortality rates from 1921 to 2023 with and without smoothing by single years of age from 0 to 94 and the last age group of 95+. From Human Mortality Database (2025), we obtain a multi-country data set used in this study for comparison. It comprises age- and sex-specific mortality rates for the 23 countries listed in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: , with the distant past shown in red and the most recent years in purple. (a) Canadian Raw Data (b) Canadian Smoothed Data (c) Japanese Raw Data (d) Japanese Smoothed Data [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Horizon-specific ranking heatmaps for the female and male mortality, measured by the MAFE and RMSFE, for the 23 countries considered [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: AUS mortality — point forecast errors (RMSFE/MAFE vs forecast horizon (h)). Country-level examples further illustrate these cross-sectional patterns. For Australia (AUS; [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: CAN mortality — point forecast errors (RMSFE/MAFE vs forecast horizon (h)). Taken together, the evidence indicates that no single fitting scheme dominates uniformly across sex, country, and horizon. The equal-weight combination serves as a robust default: it avoids the need for additional tuning, is never systematically inferior, frequently matches or exceeds the best single scheme, and attenuates end-of-h… view at source ↗
Figure 7
Figure 7. Figure 7: Horizon-specific best-method heatmaps for ASFR point forecasts, measured by RMSFE and MAFE, for the 16 countries considered [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: CAN ASFR — point forecast errors (RMSFE/MAFE vs forecast horizon (h)). Country-level evidence from Japan (JPN; [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: JPN ASFR — point forecast errors (RMSFE/MAFE vs forecast horizon (h)). Overall, no single fitting scheme dominates uniformly in fertility forecasting. Expanding is often preferred under RMSFE, while the equal-weight combined forecast frequently leads under MAFE and remains highly competitive under RMSFE. Given heterogeneous country dynamics and limited prior information on structural change, the tuning-fre… view at source ↗
Figure 10
Figure 10. Figure 10: Horizon-specific ranking heatmaps for the female and male mortality, measured by the CPD and score, for the 23 countries considered. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: AUS mortality — interval forecast evaluation (ECP / CPD / Interval Score by Horizon (h)). 20 [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: CAN mortality — interval forecast evaluation (ECP/CPD/Interval Score by Horizon (h)). For Canada ( [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Horizon-specific ranking heatmaps for the ASFR, measured by the CPD and score, for the 16 countries considered. Country-level evidence from Canada (CAN; [PITH_FULL_IMAGE:figures/full_fig_p023_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: CAN ASFR — interval forecast evaluation (ECP / CPD / Interval Score by horizon (h)). The country-level evidence from Japan (JPN; [PITH_FULL_IMAGE:figures/full_fig_p024_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: JPN ASFR — interval forecast evaluation (ECP/CPD/Interval Score by horizon (h)). The ASFR results indicate that no single scheme dominates uniformly across horizons and countries. Nevertheless, expanding most often wins on the CPD and interval score in cross-sectional summaries, consistent with fertility dynamics benefiting from a longer historical context. The equal￾weight combined method remains a robus… view at source ↗
Figure 16
Figure 16. Figure 16: Horizon-specific ranking heatmaps for the female and male mortality under the LC model, measured by the RMSFE and MAFE, for the 23 countries considered. The Australian and Canadian LC panels support the conclusion from heatmaps. Across both sexes and both accuracy measures, Rolling is the lowest or nearly lowest curve over the full horizon range. Expanding is usually the highest-error curve, and Combined … view at source ↗
Figure 17
Figure 17. Figure 17: AUS mortality — LC Point Forecast Errors (RMSFE/MAFE vs Forecast Horizon (h)) [PITH_FULL_IMAGE:figures/full_fig_p030_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: CAN mortality — LC point forecast errors (RMSFE/MAFE vs forecast horizon (h)). 30 [PITH_FULL_IMAGE:figures/full_fig_p030_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Horizon-specific ranking heatmaps for the female and male mortality under the APC model, measured by the RMSFE and MAFE, for the 23 countries considered. The Australian and Canadian APC line plots are consistent with this cross-country ranking. In both countries, Rolling has the lowest error curves for most horizons, Expanding has the highest, and Combined generally tracks the midpoint between the two sin… view at source ↗
Figure 20
Figure 20. Figure 20: AUS mortality — APC point forecast errors (RMSFE/MAFE vs forecast horizon (h)) [PITH_FULL_IMAGE:figures/full_fig_p032_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: CAN mortality — APC point forecast errors (RMSFE/MAFE vs forecast horizon (h)) 32 [PITH_FULL_IMAGE:figures/full_fig_p032_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Horizon-specific ranking heatmaps for the ASFR under the LC model, measured by the RMSFE and MAFE, for the 16 countries considered. Figures 25–27 present the corresponding APC results for ASFR. Compared with LC, the APC model produces a more horizon-dependent ranking. Rolling remains the leading method on average, winning 9.1 countries under RMSFE and 9.2 countries under MAFE, while Expanding wins 6.5 and… view at source ↗
Figure 23
Figure 23. Figure 23: CAN ASFR — LC Point Forecast Errors (RMSFE/MAFE vs Forecast Horizon (h)) [PITH_FULL_IMAGE:figures/full_fig_p034_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: JPN ASFR — LC point forecast errors (RMSFE/MAFE vs forecast horizon (h)). h = 1 and h = 2, whereas Expanding gains more wins at longer horizons. The Combined scheme has relatively few heat-map wins under APC, with average counts of fewer than one country for both metrics. 34 [PITH_FULL_IMAGE:figures/full_fig_p034_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Horizon-specific ranking heatmaps for the ASFR under the APC model, measured by the RMSFE and MAFE, for the 16 countries considered [PITH_FULL_IMAGE:figures/full_fig_p035_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: CAN ASFR — APC point forecast errors (RMSFE/MAFE vs forecast horizon (h)). 35 [PITH_FULL_IMAGE:figures/full_fig_p035_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: JPN ASFR — APC point forecast errors (RMSFE/MAFE vs forecast horizon (h)). The two country examples clarify why the APC heatmap is less uniform than the LC heatmap. For Canada, Expanding is generally more accurate at medium and long horizons, with Combined remaining close and Rolling becoming the least accurate method toward the end of the horizon range. For Japan, the opposite pattern occurs: Rolling is … view at source ↗
Figure 28
Figure 28. Figure 28: Australia and Canada mortality – LC interval forecasts: ECP by sex and forecast horizon (h) [PITH_FULL_IMAGE:figures/full_fig_p040_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: Australia and Canada mortality – APC interval forecasts: ECP by sex and forecast horizon (h). 40 [PITH_FULL_IMAGE:figures/full_fig_p040_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: Horizon-specific ranking heatmaps for the female and male mortality under the LC model, as measured by the CPD and score, for the 23 countries considered [PITH_FULL_IMAGE:figures/full_fig_p041_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: Horizon-specific ranking heatmaps for the female and male mortality under the APC model, as measured by the CPD and score, for the 23 countries considered. 41 [PITH_FULL_IMAGE:figures/full_fig_p041_31.png] view at source ↗
Figure 32
Figure 32. Figure 32: Canada and Japan ASFR – LC interval forecasts: ECP by forecast horizon (h). Figures 32–35 summarize the ASFR interval-forecast results under the LC and APC benchmarks. Under the LC specification, the heatmaps show a clear advantage for Rolling. In the ECP panel, Rolling wins on average 8.3 out of 16 countries, compared with 3.9 for Expanding and 3.8 for Combined. In the interval-score panel, Rolling again… view at source ↗
Figure 33
Figure 33. Figure 33: Canada and Japan ASFR – APC interval forecasts: ECP by forecast horizon (h) [PITH_FULL_IMAGE:figures/full_fig_p043_33.png] view at source ↗
Figure 34
Figure 34. Figure 34: Horizon-specific ranking heatmaps for the ASFR under the LC model, as measured by the CPD and score, for the 16 countries considered. 43 [PITH_FULL_IMAGE:figures/full_fig_p043_34.png] view at source ↗
Figure 35
Figure 35. Figure 35: Horizon-specific ranking heatmaps for the ASFR under the APC model, as measured by the CPD and score, for the 16 countries considered. The APC-ASFR results are more horizon-dependent. In the ECP heatmap, Rolling and Expanding are close on average, with mean win counts of 6.1 and 6.3 countries, respectively. This suggests that, for APC-based ASFR intervals, calibration alone does not identify a single domi… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

27 extracted references · 12 canonical work pages · 1 internal anchor

  1. [1]

    G., & Booth, H

    Basellini, U., Camarda, C. G., & Booth, H. (2023). Thirty years on: A review of the Lee-Carter method for forecasting mortality.International Journal of Forecasting,39(3), 1033–1049, https: //doi.org/10.1080/10920277.2000.10595882

    work page doi:10.1080/10920277.2000.10595882 2023

  2. [2]

    Booth, H. (2006). Demographic forecasting: 1980 to 2005 in review.International Journal of Forecasting, 22(3), 547–581,https://doi.org/10.1016/j.ijforecast.2006.04.001

    work page doi:10.1016/j.ijforecast.2006.04.001 2006

  3. [3]

    J., Tickle, L., & de Jong, P

    Booth, H., Hyndman, R. J., Tickle, L., & de Jong, P . (2006). Lee-Carter mortality forecasting: A multi-country comparison of variants and extensions.Demographic Research,15, 289–310, https: //doi.org/10.4054/DemRes.2006.15.9

    work page doi:10.4054/demres.2006.15.9 2006

  4. [4]

    Booth, H., Maindonald, J., & Smith, L. (2002). Applying Lee-Carter under conditions of variable mortality decline.Population Studies,56(3), 325–336,https://doi.org/10.1080/00324720215935

    work page doi:10.1080/00324720215935 2002

  5. [5]

    & Tickle, L

    Booth, H. & Tickle, L. (2008). Mortality modelling and forecasting: A review of methods.Annals of Actuarial Science,3(1-2), 3–43,https://doi.org/10.1017/S1748499500000440

    doi:10.1017/s1748499500000440 2008

  6. [6]

    Can You Trust Your Pose? Confidence Estimation in Visual Localization

    Frejka, T., Jones, G. W., & Sardon, J.-P . (2010). East Asian childbearing patterns and policy develop- ments.Population and Development Review,36(3), 579–606, https://doi.org/10.1111/j.1728-4457. 2010.00347.x

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1111/j.1728-4457 2010

  7. [7]

    & Raftery, A

    Gneiting, T. & Raftery, A. E. (2007). Strictly proper scoring rules, prediction, and estimation.Journal of the American Statistical Association: Review Article,102, 359–378, https://doi.org/10.1198/ 016214506000001437

    2007

  8. [8]

    & Renshaw, A

    Haberman, S. & Renshaw, A. (2009). On age-period-cohort parametric mortality rate projections. Insurance: Mathematics and Economics,45, 255–270, https://doi.org/10.1016/j.insmatheco.2009. 07.006

    work page doi:10.1016/j.insmatheco.2009 2009

  9. [9]

    He, X. & Ng, P . (1999). COBS: Qualitatively constrained smoothing via linear programming.Compu- tational Statistics,14, 315–337,https://doi.org/10.1007/s001800050019. Human Fertility Database (2025).Human Fertility Database. Vienna Institute of Demography (Austria), and Max Planck Institute for Demographic Research (Germany). Retrieved July 15, 2025, fro...

    work page doi:10.1007/s001800050019 1999

  10. [10]

    J., Booth, H., & Yasmeen, F

    Hyndman, R. J., Booth, H., & Yasmeen, F. (2013). Coherent mortality forecasting: the product-ratio method with functional time series models.Demography,50(1), 261–283, https://doi.org/10. 1007/s13524-012-0145-5

    2013

  11. [11]

    J., Koehler, A., Ord, J., & Snyder, R

    Hyndman, R. J., Koehler, A., Ord, J., & Snyder, R. (2008).Forecasting with Exponential Smoothing: the State-Space Approach. Berlin: Springer

    2008

  12. [12]

    Hyndman, R. J. & Shang, H. L. (2009). Forecasting functional time series.Journal of the Korean Statistical Society,38(3), 199–211,https://doi.org/10.1016/j.jkss.2009.06.002

    doi:10.1016/j.jkss.2009.06.002 2009

  13. [13]

    Hyndman, R. J. & Ullah, M. S. (2007). Robust forecasting of mortality and fertility rates: A functional data approach.Computational Statistics and Data Analysis,51(10), 4942–4956, https://doi.org/10. 1016/j.csda.2006.07.028

    2007

  14. [14]

    & Kunst, A

    Janssen, F. & Kunst, A. (2007). The choice among past trends as a basis for the prediction of future trends in old-age mortality.Population Studies,61(3), 315–326, https://doi.org/10.1080/ 00324720701571632

    2007

  15. [15]

    Lee, R. D. (1993). Modeling and forecasting the time series of US fertility: Age distribution, range, and ultimate level.International Journal of Forecasting,9(2), 187–202, https://doi.org/10.1016/ 0169-2070(93)90004-7

    1993

  16. [16]

    Lee, R. D. & Carter, L. R. (1992). Modeling and forecasting U.S. mortality.Journal of the American Statistical Association: Applications and Case Studies,87(419), 659–671, https://doi.org/10.1080/ 01621459.1992.10475265

    arXiv 1992

  17. [17]

    Maciejowska, K., Uniejewski, B., & Serafin, T. (2020). PCA forecast averaging-predicting day-ahead and intraday electricity prices.Energies,13(14), 3530,https://doi.org/10.3390/en13143530. OECD (2023).Health at a Glance 2023: OECD Indicators. Paris: OECD Publishing

    work page doi:10.3390/en13143530 2020

  18. [18]

    H., Glei, D

    Preston, S. H., Glei, D. A., & Wilmoth, J. R. (2010). Contribution of smoking to international differences in life expectancy. In E. M. Crimmins, S. H. Preston, & B. Cohen (Eds.),International Differences 47 in Mortality at Older Ages: Dimensions and Sources(pp. 105–131). Washington, D.C.: The National Academies Press

    2010

  19. [19]

    Preston, S. H. & Wang, H. (2006). Sex mortality differences in the United States: the role of cohort smoking patterns.Demography,43(4), 631–646,https://doi.org/10.1353/dem.2006.0037

    work page doi:10.1353/dem.2006.0037 2006

  20. [20]

    M., Park, H., Xie, Y., & Yeung, W.-J

    Raymo, J. M., Park, H., Xie, Y., & Yeung, W.-J. J. (2015). Marriage and family in East Asia: Continuity and change.Annual Review of Sociology,41(1), 471–492, https://doi.org/10.1146/ annurev-soc-073014-112428

    2015

  21. [21]

    Renshaw, A. E. & Haberman, S. (2006). A cohort-based extension to the Lee-Carter model for mortality reduction factors.Insurance: Mathematics and Economics,38, 556–570, https://doi.org/10.1016/j. insmatheco.2005.12.001

    doi:10.1016/j 2006

  22. [22]

    R., & Myrskyl¨a, M

    Schmertmann, C., Zagheni, E., Goldstein, J. R., & Myrskyl¨a, M. (2014). Bayesian forecasting of cohort fertility.Journal of the American Statistical Association: Applications and Case Studies,109(506), 500–513, https://doi.org/10.1080/01621459.2014.881738

    doi:10.1080/01621459.2014.881738 2014

  23. [23]

    Shang, H. L. & Haberman, S. (2024). Weighted compositional functional data analysis for modeling and forecasting life-table death counts.Journal of Forecasting,43(8), 3051–3071, https://doi.org/ 10.1002/for.3171

    work page doi:10.1002/for.3171 2024

  24. [24]

    Shang, H. L. & Haberman, S. (2025). Constructing prediction intervals for the age distribution of deaths.Scandinavian Actuarial Journal,in press, https://doi.org/10.1080/03461238.2025. 2544265

    doi:10.1080/03461238.2025 2025

  25. [25]

    Shang, H. L. & Xu, R. (2022). Change point detection for COVID-19 excess deaths in Belgium.Journal of Population Research,39(4), 557–565,https://doi.org/10.1007/s12546-021-09256-2

    work page doi:10.1007/s12546-021-09256-2 2022

  26. [26]

    Wensink, M., Alvarez, J.-A., Rizzi, S., Janssen, F., & Lindahl-Jacobsen, R. (2020). Progression of the smoking epidemic in high-income regions and its effects on male-female survival differences: A cohort-by-age analysis of 17 countries.BMC Public Health,20(1), 39, https://doi.org/10.1186/ s12889-020-8148-4

    2020

  27. [27]

    Wood, S. N. (1994). Monotonic smoothing splines fitted by cross validation.SIAM Journal on Scientific Computing,15(5), 1126–1133,https://doi.org/10.1137/0915069. 48

    work page doi:10.1137/0915069 1994