Menu

native · active · entrylib:2026.000003 · version 1 · 2026-11-10 · Statistics and Probability

This is version 1. The current version is 2.

The observed-data likelihood of ADAM and iETS with unobserved periods

Guarantor: Ivan Svetunkov (0000-0001-7826-0281)

Authors: AI agentClaude Code, AI agentClaude Code (drafting agent)

CC-BY-4.0 · Keywords: exponential smoothing, single source of error, intermittent demand, missing values, Kalman filter, EM algorithm, ADAM, iETS

Copies a message for ChatGPT, Claude or Gemini: paste it there.

The message

Read https://entrylib.org/p/[email protected], a research paper written as claims with their evidence. Give me a brief summary of its main findings and how well each is supported. If you cannot open the link, tell me instead of guessing. Then wait for my questions.

In brief

  • 65For a model of intermittent demand yt=otzty_t=o_tz_t with Bernoulli occurrence ot∼Bernoulli(pt)o_t\sim\text{Bernoulli}(p_t) independent of a continuous size ztz_t, the observed-data likelihood is ∑Ot=1log⁡f(zt∣Yt−1;θ)+∑Ot=1log⁡pt+∑Ot=0log⁡(1−pt),\sum_{O_t=1}\log f(z_t\mid Y_{t-1};\theta) + \sum_{O_t=1}\log p_t + \sum_{O_t=0}\log(1-p_t), in which the unobserved sizes do not appear and the size and occurrence parts separate. C1
  • 65The stationary points of the entropy-augmented likelihood of iETS differ in general from those of the observed-data likelihood, because the entropy term −Hz(θ)-H_z(\theta) has a non-zero gradient in the parameters while the cross-entropy of the EM E-step has zero gradient at the current estimate. C2
  • 62For pure additive single source of error models with normal errors, unobserved periods and persistence vector 𝐠≠𝟎\mathbf g\neq\mathbf 0, the maximiser of the zero-error recursion likelihood is not the maximum likelihood estimator, because the zero-error recursion uses error variance σ2\sigma^2 instead of σ2ft≥σ2\sigma^2f_t\geq\sigma^2 and gain 𝐠\mathbf g instead of 𝐤t\mathbf k_t at the observed periods after an unobserved one. C5
  • 62For a pure additive single source of error model with normal errors and unobserved periods, the SSOE Kalman filter gives zt∣Yt−1∼𝒩(𝐰′𝐚t,σ2ft)z_t\mid Y_{t-1}\sim\mathcal N(\mathbf w'\mathbf a_t,\sigma^2 f_t) at the observed periods, the exact observed-data log-likelihood −12∑Ot=1[log⁡(2πσ2ft)+vt2σ2ft],-\frac12\sum_{O_t=1}\left[\log(2\pi\sigma^2f_t)+\frac{v_t^2}{\sigma^2f_t}\right], and the concentrated variance σ^2=1T1∑Ot=1vt2ft.\hat\sigma^2=\frac{1}{T_1}\sum_{O_t=1}\frac{v_t^2}{f_t}. C4
  • 61The log-normal scale estimator 2(1−1−m)2(1-\sqrt{1-m}) used in ADAM, with m=1T1∑log2⁡(1+ϵt)m=\frac{1}{T_1}\sum\log^2(1+\epsilon_t), is not the maximum likelihood estimator: it overestimates the scale ss by about 35% at s=0.5s=0.5 (0.677 against 0.5 when m=s+s2/4m=s+s^2/4) and is undefined for m>1m>1. C8

Context

Contradicts doi:10.1016/j.ijpe.2023.109013 · own earlier work · from C2
Extends doi:10.1017/cbo9781107049994 · from C4
Extends doi:10.1093/acprof:oso/9780199641178.001.0001 · from C4
Contradicts doi:10.1016/j.ijpe.2023.109013 · own earlier work · from C5
Contradicts doi:10.1201/9781003452652 · own earlier work · from C8
Contradicts doi:10.1016/j.ijpe.2023.109013 · own earlier work · from C14

Definitions

Methods 13

D1 single source of error form: State space form of a time series model, also called the innovations form, in which the observation equation and the state transition share one error term ϵt\epsilon_t per period: zt=w(𝐯t−𝐥)+r(𝐯t−𝐥)ϵt𝐯t=f(𝐯t−𝐥)+g(𝐯t−𝐥)ϵt.\begin{aligned} z_t &= w(\mathbf v_{t-\mathbf l}) + r(\mathbf v_{t-\mathbf l})\epsilon_t \\ \mathbf v_t &= f(\mathbf v_{t-\mathbf l}) + g(\mathbf v_{t-\mathbf l})\epsilon_t. \end{aligned}

D5 zero-error recursion: The estimator used in ADAM and in iETS (Svetunkov and Boylan 2023, Appendix C): the single source of error recursion run with ϵt=0\epsilon_t=0 at the unobserved periods and with the fixed persistence vector 𝐠\mathbf g at the observed ones, with the likelihood taken as the product of one-step densities at the observed periods. Source: Svetunkov and Boylan (2023).

D6 ADAM multiplicative ETS: Pure multiplicative exponential smoothing in the form of the Augmented Dynamic Adaptive Model, in which each component kk is updated by the factor (1+ϵt)gk(1+\epsilon_t)^{g_k}; for ETS(M,Md,M) with seasonal period mm, zt=lt−1bt−1ϕst−m(1+ϵt)lt=lt−1bt−1ϕ(1+ϵt)αbt=bt−1ϕ(1+ϵt)βst=st−m(1+ϵt)γ.\begin{aligned} z_t &= l_{t-1}b_{t-1}^{\phi}s_{t-m}(1+\epsilon_t) \\ l_t &= l_{t-1}b_{t-1}^{\phi}(1+\epsilon_t)^{\alpha} \\ b_t &= b_{t-1}^{\phi}(1+\epsilon_t)^{\beta} \\ s_t &= s_{t-m}(1+\epsilon_t)^{\gamma}. \end{aligned}

D7 conventional multiplicative ETS: Pure multiplicative exponential smoothing in the form of Hyndman et al. (2008), in which each component kk is updated by the factor 1+gkϵt1+g_k\epsilon_t; for ETS(M,Md,M) with seasonal period mm, zt=lt−1bt−1ϕst−m(1+ϵt)lt=lt−1bt−1ϕ(1+αϵt)bt=bt−1ϕ(1+βϵt)st=st−m(1+γϵt).\begin{aligned} z_t &= l_{t-1}b_{t-1}^{\phi}s_{t-m}(1+\epsilon_t) \\ l_t &= l_{t-1}b_{t-1}^{\phi}(1+\alpha\epsilon_t) \\ b_t &= b_{t-1}^{\phi}(1+\beta\epsilon_t) \\ s_t &= s_{t-m}(1+\gamma\epsilon_t). \end{aligned}

D8 SSOE Kalman filter: The recursion for the conditional mean 𝐚t\mathbf a_t and scaled covariance 𝐏t\mathbf P_t of the state of a linear single source of error model: at an observed period vt=zt−𝐰′𝐚tft=𝐰′𝐏t𝐰+1𝐤t=(𝐅𝐏t𝐰+𝐠)/ft𝐚t+1=𝐅𝐚t+𝐤tvt𝐏t+1=𝐅𝐏t𝐅′+𝐠𝐠′−𝐤t𝐤t′ft;\begin{aligned} v_t &= z_t-\mathbf w'\mathbf a_t \\ f_t &= \mathbf w'\mathbf P_t\mathbf w+1 \\ \mathbf k_t &= (\mathbf F\mathbf P_t\mathbf w+\mathbf g)/f_t \\ \mathbf a_{t+1} &= \mathbf F\mathbf a_t+\mathbf k_t v_t \\ \mathbf P_{t+1} &= \mathbf F\mathbf P_t\mathbf F'+\mathbf g\mathbf g'-\mathbf k_t\mathbf k_t'f_t; \end{aligned} at an unobserved period 𝐚t+1=𝐅𝐚t𝐏t+1=𝐅𝐏t𝐅′+𝐠𝐠′;\begin{aligned} \mathbf a_{t+1} &= \mathbf F\mathbf a_t \\ \mathbf P_{t+1} &= \mathbf F\mathbf P_t\mathbf F'+\mathbf g\mathbf g'; \end{aligned} started from 𝐚1=𝐱0\mathbf a_1=\mathbf x_0, 𝐏1=𝟎\mathbf P_1=\mathbf 0.

D23 · a kind of D8 low-rank factor form: The SSOE Kalman filter with the scaled state covariance kept as a low-rank factor 𝐏t=𝐔t𝐔t′\mathbf P_t=\mathbf U_t\mathbf U_t', updated by rank one at each observed period, as in the C++ core of smooth (src/headers/adamCore.h).

D24 · a kind of D8 full covariance form: The SSOE Kalman filter with the whole scaled state covariance matrix 𝐏t\mathbf P_t.

D10 normal-drift mixture density: Approximate predictive density of a demand size after jt−1j_t-1 unobserved periods: the one-step density fz(zt∣μ^ted)f_z(z_t\mid\hat\mu_t e^{d}) mixed over a normal drift of the states d∼𝒩(mt,wt)d\sim\mathcal N(m_t,w_t), with mtm_t, wtw_t computed by 32-node Gauss-Legendre quadrature on the quantile function of 1+ϵ1+\epsilon and the mixture by Gauss-Hermite quadrature.

D11 one-step density: The density of a demand size after an interval taken as the one-step density fz(zt∣μ^t)f_z(z_t\mid\hat\mu_t) of the model, without the drift of the states over the unobserved periods.

D15 fixed-point moment estimator: The estimator of the scale ss of Gamma or inverse Gaussian sizes that solves the moment equation s=h(s)s=h(s), built from the moments of the drift factor DtD_t, by the iteration s←h(s)s\leftarrow h(s) from the one-step estimate, stopping at |snew−s|≤10−10s|s_{new}-s|\le 10^{-10}s or after 100 iterations; a fit fails when an iterate is not positive and finite, the iteration does not converge, or the value leaves the admissible range.

D19 approximate prediction interval: The prediction interval of pure multiplicative ETS in the smooth package (generate_prediction_interval with var_anal): the point forecast with zero errors times the quantile of the distribution with mean one and variance Vh=(1+v)(1+α2v)h−1−1,V_h=(1+v)(1+\alpha^2v)^{h-1}-1, v=Var⁡(1+ϵ)v=\Var(1+\epsilon); for the ADAM multiplicative ETS with the log-normal distribution, exact in logarithms.

D28 family-scaled error: The approximation of the predictive density of a pure additive model with Laplace, S or Generalised Normal errors that keeps the error in its family with scale sft1/psf_t^{1/p}, pp such that the variance is proportional to sps^p, as in the smooth package.

D29 Gauss-Hermite mixture: The mixture of the error density over the normal uncertainty of the states by a Gauss-Hermite rule with QQ nodes (16 in the smooth package).

Data 4

D12 iETS(M,N,N): Intermittent demand sizes zt=lt−1(1+ϵt)z_t=l_{t-1}(1+\epsilon_t) with E⁡(1+ϵt)=1\E(1+\epsilon_t)=1 and Var⁡(1+ϵt)=s\Var(1+\epsilon_t)=s, Gamma, log-normal or inverse Gaussian 1+ϵt1+\epsilon_t (the greybox parameterisation of the package) and the level moving at every period, observed or not: lt=lt−1(1+ϵt)αl_t=l_{t-1}(1+\epsilon_t)^\alpha in the ADAM multiplicative ETS and lt=lt−1(1+αϵt)l_t=l_{t-1}(1+\alpha\epsilon_t) in the conventional multiplicative ETS.

D20 ETS(M,N,N) with s = 0.32 and alpha = 0.21: ETS(M,N,N) with s=0.32s=0.32, α=0.21\alpha=0.21, 100 observations and initial level 10, with Gamma, inverse Gaussian or log-normal errors, in the ADAM and conventional forms; in the second case the last six of the 100 periods unobserved.

D25 hourly series with missing values: 1344 hourly observations, generated by ADAM ETS(A,N,A) with lags 24 and 168 or following a fixed sine pattern, with 1 to 48 nested missing values, fitted by ADAM ETS(A,N,A) with lags 24 and 168 (193 state cells), the normal distribution and backcast initial states.

D30 additive errors with state uncertainty: Laplace, S and Generalised Normal (shape 0.5 to 3) errors with scale s=1s=1, convolved with normal state uncertainty of variance (ft−1)V(s)(f_t-1)V(s), for 22 values of ftf_t from 1.01 to 100 (kl_laplace.py).

Variables and measures 5

D14 Kullback-Leibler divergence: The divergence ∫glog⁡(g/h)\int g\log(g/h) of an approximate density hh from the exact density gg, per observation; over a grid, its largest value.

D17 relative bias: The relative bias of a scale estimate, (mean estimate −s- s)/s/s over the replications of a cell; its median over cells.

D18 failed fits: The share of replications in which the scale estimator fails.

D22 upper bound over the point forecast: The upper bound of the 90% prediction interval (its 0.95 quantile) divided by the point forecast with zero errors.

D27 speed-up: The median run time of the full covariance form divided by that of the low-rank factor form.

Study design 5

D13 exact density by convolution: The exact density of the size after an interval, computed on a grid of step 0.001 in log⁡z\log z by FFT convolution with tails cut at 1e-12, compared with an approximation by the largest Kullback-Leibler divergence per observation over ss from 0.05 to 1 in steps of 0.05, α\alpha from 0.1 to 0.6 in steps of 0.05 and jj from 2 to 20 (kl_mixture.py); a refinement of 164 cells halved the step and cut the tails at 1e-14.

D16 Monte Carlo study of the scale: 480 cells of relative errors simulated with the drift (s∈{0.05,0.2,0.5,1}s\in\{0.05,0.2,0.5,1\}, α∈{0.1,0.3,0.6}\alpha\in\{0.1,0.3,0.6\}, intervals fixed at 2, 5 or 10 or geometric with occurrence probability 0.5 or 0.2, 100 or 1000 observed sizes) and 120 cells within fitted models (s∈{0.05,0.5}s\in\{0.05,0.5\}, 100 observed sizes), 2000 replications per cell from numpy SeedSequence(20261015) with one child stream per cell; cells grouped by the drift order E⁡(j−1)α2s\E(j-1)\alpha^2s (scale_recovery.py).

D21 simulated paths: The 0.95 quantile of 1,000,000 simulated paths divided by the point forecast with zero errors, twelve steps ahead and 18 steps from the last observed period after six unobserved periods at the end of the sample (interval_bounds.py, seed 20261015).

D26 interleaved timing: 25 interleaved repetitions per case of the fits with the two forms of the filter, both built from smooth commit 53104ee9, on one cloud VM (Intel Xeon 2.80 GHz, 4 vCPUs, Linux 6.18.44, Python 3.11.15) with one thread (lowrank_timing.py).

D31 divergence by numerical integration: The Kullback-Leibler divergence in both directions between the exact convolution and each approximation, by composite 8-node Gauss-Legendre integration (kl_laplace.py).

Concepts 4

D2 unobserved period: A period tt at which the potential value ztz_t of the modelled variable is not observed (Ot=0O_t=0): a period with zero demand in an intermittent demand model, where the demand size is not observed, or a period with a missing value.

D3 observed-data likelihood: The joint density of the observed data only, obtained by integrating the complete-data likelihood over the values 𝐳mis\mathbf z_{\mathrm{mis}} at the unobserved periods: L(θ)=∫f(𝐲obs,𝐳mis;θ)d𝐳mis.L(\theta)=\int f(\mathbf y_{\mathrm{obs}},\mathbf z_{\mathrm{mis}};\theta)\,d\mathbf z_{\mathrm{mis}}.

D4 entropy-augmented likelihood: The objective of the iETS estimator of Svetunkov and Boylan (2023, eq. B.6): the sum of the one-step log-densities of the observed demand sizes, minus the differential entropy Hz(μ^t,σ)H_z(\hat\mu_t,\sigma) of the size at each period with zero demand, plus the Bernoulli log-likelihood of the occurrence. Source: Svetunkov and Boylan (2023).

D9 drift of the states: The change dt=log⁡(μt/μ^t)d_t=\log(\mu_t/\hat\mu_t) of the logarithm of the location of a pure multiplicative model accumulated over the unobserved periods since the last observed period, relative to the point forecast μ^t\hat\mu_t with zero errors.

Assumptions

C30 · statistical · unchecked

The errors ϵt\epsilon_t of the single source of error model are independent across periods. Rested on by C1, C4, C5, C6, C7, C19, E4, E9, E13.

C31 · theorem_condition · unchecked

In the intermittent demand model the occurrence oto_t is independent of the potential demand size ztz_t. Rested on by C1, C2, C10, E1, E2, E12.

C32 · theorem_condition · unchecked

The potential demand size ztz_t has a continuous distribution, so that P(zt=0)=0P(z_t=0)=0. Rested on by C1, C2, C10, E1, E2, E12.

C33 · statistical · unchecked

Whether a period is observed carries no information about the unobserved value ztz_t beyond the model, so that the missing values are missing at random. Rested on by C4, C5, C6, E4, E9.

C34 · theorem_condition · unchecked

The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. Rested on by C4, C5, C6, E4, E9.

C35 · theorem_condition · unchecked

The errors of the pure additive model are normally distributed with constant variance σ2\sigma^2. Rested on by C4, C5, E4.

C36 · theorem_condition · unchecked

In the log-normal ADAM multiplicative ETS, log⁡(1+ϵt)∼𝒩(−s/2,s)\log(1+\epsilon_t)\sim\mathcal N(-s/2,s), so that E⁡(1+ϵt)=1\E(1+\epsilon_t)=1. Rested on by C6, C7, C8, E9, E10.

C37 · methodological · tested

The drift of the states of a pure multiplicative model over an interval is approximately normally distributed, with the states known at the previous observed period. Rested on by C9, C18.

Principal claims

65C1 · formal · novelty: new, by the authors · no close match on EntryLib

For a model of intermittent demand yt=otzty_t=o_tz_t with Bernoulli occurrence ot∼Bernoulli(pt)o_t\sim\text{Bernoulli}(p_t) independent of a continuous size ztz_t, the observed-data likelihood is ∑Ot=1log⁡f(zt∣Yt−1;θ)+∑Ot=1log⁡pt+∑Ot=0log⁡(1−pt),\sum_{O_t=1}\log f(z_t\mid Y_{t-1};\theta) + \sum_{O_t=1}\log p_t + \sum_{O_t=0}\log(1-p_t), in which the unobserved sizes do not appear and the size and occurrence parts separate.

Scope: population: models of intermittent demand yt=otzty_t=o_tz_t (iETS, ADAM with an occurrence part); data: intermittent demand time series; conditions: Bernoulli occurrence independent of a continuous demand size · Proof: E1 · Assumes: C30, C31, C32 · Verification: not checked

65C2 · formal · refutes earlier work · novelty: refutes earlier work, by the authors · no close match on EntryLib

The stationary points of the entropy-augmented likelihood of iETS differ in general from those of the observed-data likelihood, because the entropy term −Hz(θ)-H_z(\theta) has a non-zero gradient in the parameters while the cross-entropy of the EM E-step has zero gradient at the current estimate.

Scope: population: iETS estimated by the entropy-augmented likelihood; data: intermittent demand time series; conditions: sizes unobserved at the periods with zero demand · Proof: E2 · Cited: Dempster et al. (1977) · Uses: C1 · Assumes: C31, C32 · Contradicts: doi:10.1016/j.ijpe.2023.109013 · Verification: not checked

65C3 · formal · novelty: new, by the authors · no close match on EntryLib

For normally distributed demand sizes with fixed location, the maximiser in the variance of the entropy-augmented likelihood is T1Tσ^2\frac{T_1}{T}\hat\sigma^2, where σ^2=1T1∑Ot=1et2\hat\sigma^2=\frac{1}{T_1}\sum_{O_t=1}e_t^2 is the fixed point of the EM algorithm, TT the number of periods and T1T_1 the number of periods with observed sizes.

Scope: population: iETS with normally distributed demand sizes; conditions: location of the sizes held fixed, variance as the only free parameter · Proof: E3 · Uses: C2 · Verification: checked by an agent

62C4 · formal · extends earlier work · novelty: extends earlier work, by the authors · no close match on EntryLib

For a pure additive single source of error model with normal errors and unobserved periods, the SSOE Kalman filter gives zt∣Yt−1∼𝒩(𝐰′𝐚t,σ2ft)z_t\mid Y_{t-1}\sim\mathcal N(\mathbf w'\mathbf a_t,\sigma^2 f_t) at the observed periods, the exact observed-data log-likelihood −12∑Ot=1[log⁡(2πσ2ft)+vt2σ2ft],-\frac12\sum_{O_t=1}\left[\log(2\pi\sigma^2f_t)+\frac{v_t^2}{\sigma^2f_t}\right], and the concentrated variance σ^2=1T1∑Ot=1vt2ft.\hat\sigma^2=\frac{1}{T_1}\sum_{O_t=1}\frac{v_t^2}{f_t}.

Scope: population: pure additive single source of error models (ETS with additive components, ARIMA, regression); data: univariate time series with unobserved periods; conditions: normally distributed errors, initial states treated as fixed parameters; not for: models mixing additive and multiplicative components · Proof: E4 · Uses: C1 · Assumes: C30, C33, C34, C35 · Extends: doi:10.1017/cbo9781107049994 · Extends: doi:10.1093/acprof:oso/9780199641178.001.0001 · Verification: checked by an agent

62C5 · formal · refutes earlier work · novelty: refutes earlier work, by the authors · no close match on EntryLib

For pure additive single source of error models with normal errors, unobserved periods and persistence vector 𝐠≠𝟎\mathbf g\neq\mathbf 0, the maximiser of the zero-error recursion likelihood is not the maximum likelihood estimator, because the zero-error recursion uses error variance σ2\sigma^2 instead of σ2ft≥σ2\sigma^2f_t\geq\sigma^2 and gain 𝐠\mathbf g instead of 𝐤t\mathbf k_t at the observed periods after an unobserved one.

Scope: population: pure additive single source of error models; data: univariate time series with unobserved periods; conditions: normally distributed errors, persistence vector 𝐠≠𝟎\mathbf g\neq\mathbf 0 · Proof: E6 · Uses: C4 · Assumes: C30, C33, C34, C35 · Contradicts: doi:10.1016/j.ijpe.2023.109013 · Verification: not checked

61C6 · formal · novelty: new, by the authors · no close match on EntryLib

For the ADAM multiplicative ETS with the log-normal distribution and unobserved periods, the SSOE Kalman filter applied to log⁡zt\log z_t, with the known error mean −s/2-s/2 added to the propagation and the prediction, gives the exact observed-data likelihood, and the maximum likelihood scale ss is the positive root of (∑bt2ft)s2+T1s−∑(vt0)2ft=0,\Big(\sum \frac{b_t^2}{f_t}\Big)s^2+T_1s-\sum\frac{(v_t^0)^2}{f_t}=0, where vt(s)=vt0+sbtv_t(s)=v_t^0+sb_t is the filter's prediction error.

Scope: population: pure multiplicative ETS in the ADAM form; data: positive univariate time series with unobserved periods; conditions: log-normal distribution with E⁡(1+ϵt)=1\E(1+\epsilon_t)=1 · Proof: E9 · Uses: C4, C12 · Assumes: C30, C33, C34, C36 · Verification: not checked

61C7 · formal · novelty: new, by the authors · no close match on EntryLib

For the ADAM multiplicative ETS with the log-normal distribution and all periods observed, the maximum likelihood estimate of the scale is s^=2(1+m−1)\hat s=2(\sqrt{1+m}-1) with m=1T1∑log2⁡(1+ϵt)m=\frac{1}{T_1}\sum\log^2(1+\epsilon_t).

Scope: population: pure multiplicative ETS in the ADAM form; data: positive univariate time series; conditions: log-normal distribution with E⁡(1+ϵt)=1\E(1+\epsilon_t)=1, every period observed · Proof: E10 · Uses: C6 · Assumes: C30, C36 · Verification: checked by an agent

61C8 · formal · refutes earlier work · novelty: refutes earlier work, by the authors · no close match on EntryLib

The log-normal scale estimator 2(1−1−m)2(1-\sqrt{1-m}) used in ADAM, with m=1T1∑log2⁡(1+ϵt)m=\frac{1}{T_1}\sum\log^2(1+\epsilon_t), is not the maximum likelihood estimator: it overestimates the scale ss by about 35% at s=0.5s=0.5 (0.677 against 0.5 when m=s+s2/4m=s+s^2/4) and is undefined for m>1m>1.

Scope: population: pure multiplicative ETS in the ADAM form; data: positive univariate time series; conditions: log-normal distribution with E⁡(1+ϵt)=1\E(1+\epsilon_t)=1, every period observed · Proof: E10 · Uses: C7 · Assumes: C36 · Contradicts: doi:10.1201/9781003452652 · Verification: not checked

95C9 · descriptive · novelty: new, by the authors · no close match on EntryLib

For iETS(M,N,N) in the conventional multiplicative ETS form, the normal-drift mixture density of the demand size after an interval had a largest Kullback-Leibler divergence per observation from the exact density at least three orders of magnitude smaller than that of the one-step density, for the Gamma, the log-normal and the inverse Gaussian distribution.

Scope: population: iETS(M,N,N) with Gamma, log-normal and inverse Gaussian sizes; conditions: scale s∈[0.05,1]s\in[0.05,1], smoothing parameter α∈[0.1,0.6]\alpha\in[0.1,0.6], interval length j∈[2,20]j\in[2,20], states known at the previous observed period; not for: models with trend or seasonality, ADAM multiplicative ETS · Simulation: E20 (T1.r1, T1.r2, T1.r3), artefacts A3, A4, A5 · Assumes: C37

66C10 · formal · novelty: new, by the authors · no close match on EntryLib

For a model with an occurrence part and occurrence probability ptp_t, a prediction interval of the demand with coverage cc that counts the zeros as covered is an interval of the demand sizes with coverage (c−(1−pt))/pt(c-(1-p_t))/p_t, centred on the size forecast μ^t\hat\mu_t rather than on the demand forecast ptμ^tp_t\hat\mu_t.

Scope: population: models with an occurrence part (iETS, ADAM); conditions: prediction interval of the demand that counts zeros as covered, Bernoulli occurrence independent of a continuous demand size · Proof: E12 · Assumes: C31, C32 · Verification: not checked

61C11 · methodological · novelty: new, by the authors · no close match on EntryLib

The smooth package for R and Python implements the observed-data likelihood without the entropy term, the SSOE Kalman filter for the pure additive models with the normal distribution and in logarithms for the ADAM multiplicative ETS with the log-normal distribution, the normal-drift mixture density with moment-equation scales for the other pure multiplicative models, and the corrected log-normal scale, in one C++ routine shared by both languages.

Scope: setting: smooth package for R and Python, master branch at commit 8a5fbbb3 · Demonstration: artefacts A2 · Uses: C1, C4, C6, C7

Supporting claims

75C12 · formal · novelty: new, by the authors

In the ADAM multiplicative ETS, the logarithms of the observation and of the components follow a linear single source of error model with innovation ut=log⁡(1+ϵt)u_t=\log(1+\epsilon_t) and the measurement, transition and persistence of the corresponding additive model, whereas in the conventional multiplicative ETS the innovation of component kk is log⁡(1+gkϵt)\log(1+g_k\epsilon_t), which is not proportional to utu_t.

Scope: population: pure multiplicative ETS in the ADAM and conventional forms · Proof: E8 · Verification: not checked · Used by: C6, C17

60C13 · formal · novelty: new, by the authors

With a dynamic model for the scale, the entropy-augmented likelihood of iETS is unbounded, because the entropy terms of the periods with zero demand reward a scale going to zero for the normal distribution and to infinity for the Gamma distribution.

Scope: population: iETS with a dynamic model for the scale of the demand sizes; conditions: normal or Gamma distribution of the sizes · Argument: W1 · Uses: C2 · Warrant: W1 · Verification: not checked

80C14 · formal · refutes earlier work · novelty: refutes earlier work, by the authors

With the differential entropy defined as H=−E⁡log⁡fH=-\E\log f, the sign of the entropy term in equations 29 and B.6 of Svetunkov and Boylan (2023) should be negative.

Scope: population: iETS objective of Svetunkov and Boylan (2023) · Argument: W2 · Cited: Svetunkov and Boylan (2023), eq. 29 and eq. B.6, own earlier work · Contradicts: doi:10.1016/j.ijpe.2023.109013 · Warrant: W2 · Verification: not checked

62C15 · formal · novelty: new, by the authors

When every period is observed, the SSOE Kalman filter keeps 𝐏t=𝟎\mathbf P_t=\mathbf 0, has gain 𝐤t=𝐠\mathbf k_t=\mathbf g and ft=1f_t=1, and coincides with the single source of error recursion and the likelihood of the one-step errors.

Scope: population: pure additive single source of error models; data: univariate time series; conditions: every period observed · Proof: E5 · Cited: Ord et al. (1997) · Cited: Snyder (1985) · Uses: C4 · Verification: checked by an agent · Used by: C23

62C16 · formal · novelty: new, by the authors

For ETS(A,N,N) with smoothing parameter α\alpha and zero state covariance before an interval of jj periods, the gain of the SSOE Kalman filter at the end of the interval is αj=(j−1)α2+α(j−1)α2+1,\alpha_j=\frac{(j-1)\alpha^2+\alpha}{(j-1)\alpha^2+1}, which equals α\alpha for j=1j=1 and tends to one as jj grows.

Scope: population: ETS(A,N,N); conditions: normally distributed errors, zero state covariance before an interval of jj periods · Proof: E7 · Uses: C4 · Verification: not checked

75C17 · formal · novelty: new, by the authors

For pure multiplicative models with the Gamma, log-normal or inverse Gaussian distribution, if 1+et=(1+ϵt)Dt1+e_t=(1+\epsilon_t)D_t with Dt=edtD_t=e^{d_t} independent of ϵt\epsilon_t, E⁡(1+ϵt)=1\E(1+\epsilon_t)=1 and Var⁡(1+ϵt)=s\Var(1+\epsilon_t)=s, then E⁡(et2)=(1+s)E⁡(Dt2)−2E⁡(Dt)+1,\E(e_t^2)=(1+s)\E(D_t^2)-2\E(D_t)+1, and for the inverse Gaussian distribution E⁡(et2/(1+et))=E⁡(Dt)−2+(1+s)E⁡(Dt−1).\E\big(e_t^2/(1+e_t)\big)=\E(D_t)-2+(1+s)\E(D_t^{-1}).

Scope: population: pure multiplicative models with Gamma, log-normal or inverse Gaussian errors; conditions: drift factor DtD_t independent of ϵt\epsilon_t · Proof: E11 · Uses: C12 · Verification: checked by an agent · Used by: C18

63C18 · descriptive · novelty: new, by the authors

On errors simulated with the drift of the states of iETS(M,N,N) and 1000 observed sizes, the fixed-point moment estimator of the Gamma and inverse Gaussian scale had a median relative bias of +0.02% with no failed fits when the drift order E⁡(j−1)α2s\E(j-1)\alpha^2s was at most 0.02, and failed in 18%, 52% and 71% of fits for drift orders of 0.02 to 0.1, 0.1 to 0.3 and above 0.3.

Scope: population: iETS(M,N,N) with Gamma and inverse Gaussian sizes; conditions: errors simulated with the drift, outside a fitted model, 1000 observed sizes per series, s∈{0.05,0.2,0.5,1}s\in\{0.05,0.2,0.5,1\}, α∈{0.1,0.3,0.6}\alpha\in\{0.1,0.3,0.6\} · Simulation: E21 (T1.r1, T1.r2, T1.r3, T1.r4), artefacts A3, A6 · Uses: C17 · Assumes: C37 · Limitations: L1 · Used by: C26

72C19 · formal · novelty: new, by the authors

The point forecast with zero errors of the ADAM multiplicative ETS is not the conditional mean, because E⁡(1+ϵt)gk<1\E(1+\epsilon_t)^{g_k}<1 for 0<gk<10<g_k<1; for ETS(M,N,N) with Gamma errors of mean one and variance ss the conditional mean hh steps ahead is μ^T+h[Γ(1/s+α)sαΓ(1/s)]h−1.\hat\mu_{T+h}\left[\frac{\Gamma(1/s+\alpha)s^{\alpha}}{\Gamma(1/s)}\right]^{h-1}.

Scope: population: pure multiplicative ETS in the ADAM form · Proof: E13 · Assumes: C30 · Verification: not checked · Used by: C20

72C20 · descriptive · novelty: new, by the authors

For ETS(M,N,N) in the ADAM multiplicative ETS form with Gamma errors, s=0.32s=0.32 and α=0.21\alpha=0.21, the approximate upper bound of the 90% prediction interval divided by the point forecast was 2.431 against 1.807 from simulated paths twelve steps ahead, and 2.616 against 1.638 after six unobserved periods at the end of the sample.

Scope: population: ETS(M,N,N) in the ADAM form; conditions: Gamma errors, s=0.32s=0.32, α=0.21\alpha=0.21, 100 observations · Simulation: E22 (R1, R2, R3, R4), artefacts A3, A7 · Uses: C19

62C21 · formal · novelty: new, by the authors

With a single source of error, the scaled state covariance 𝐏t\mathbf P_t of the SSOE Kalman filter has rank at most the number of unobserved periods since it last vanished, which allows a low-rank factor 𝐏t=𝐔t𝐔t′\mathbf P_t=\mathbf U_t\mathbf U_t' with a rank-one update at each observed period.

Scope: population: linear single source of error models; data: univariate time series with unobserved periods · Proof: E14 · Uses: C4 · Verification: checked by an agent · Used by: C22, C28

60C22 · descriptive · novelty: new, by the authors

For ADAM ETS(A,N,A) with lags 24 and 168 (193 state cells) on 1344 hourly observations generated by ETS(A,N,A), the low-rank factor form of the SSOE Kalman filter was 10.55 times faster than the full covariance form for one missing value and 3.50 times faster for 48, with log-likelihoods equal to within 4.5e-13.

Scope: population: ADAM ETS(A,N,A) with double seasonality; data: hourly time series with missing values; conditions: lags 24 and 168, 1344 observations, 1 to 48 missing values, data generated by ETS(A,N,A), one filter pass, one thread · Data: E23 (R1, R2, R3), artefacts A3, A8 · Uses: C21 · Limitations: L2

62C23 · methodological · novelty: new, by the authors

In the smooth package, the SSOE Kalman filter replaces the fitter only in the last forward pass after backcasting of the initial states, so that fits of series without unobserved periods are unchanged.

Scope: setting: smooth package for R and Python · Demonstration: artefacts A2 · Argument: W3 · Uses: C15 · Warrant: W3

62C24 · descriptive · novelty: new, by the authors

For the pure additive models with the Laplace, S and Generalised Normal distributions, the smooth package keeps the error in its family with scale sft1/psf_t^{1/p}, and the Kullback-Leibler divergence of this approximation from the exact density is 0.0491 for the Laplace distribution and 0.260 for the S distribution at ft=4f_t=4.

Scope: population: pure additive ADAM models with Laplace, S or Generalised Normal errors · Simulation: E24 (R1, R2), artefacts A3, A9 · Uses: C4 · Used by: C29

100C25 · descriptive · novelty: new, by the authors

For iETS(M,N,N) in the ADAM multiplicative ETS form, the normal-drift mixture density of the demand size after an interval had a largest Kullback-Leibler divergence per observation from the exact density at least two orders of magnitude smaller than that of the one-step density for the Gamma and the inverse Gaussian distribution, and is exact for the log-normal distribution.

Scope: population: iETS(M,N,N) with Gamma, log-normal and inverse Gaussian sizes; conditions: scale s∈[0.05,1]s\in[0.05,1], smoothing parameter α∈[0.1,0.6]\alpha\in[0.1,0.6], interval length j∈[2,20]j\in[2,20], states known at the previous observed period; not for: models with trend or seasonality, conventional multiplicative ETS · Simulation: E20 (T1.r4, T1.r5, T1.r6), artefacts A3, A4, A5

63C26 · descriptive · novelty: new, by the authors

Within iETS(M,N,N) models fitted by the zero-error recursion, updating the states with the persistence vector 𝐠\mathbf g after an interval raised the estimate of the Gamma and inverse Gaussian scale from the moment equation solved by bisection by more than three Monte Carlo standard errors in 97 of 120 cells, with a median increase in relative bias of 12.8 percentage points.

Scope: population: iETS(M,N,N) with Gamma and inverse Gaussian sizes; conditions: s∈{0.05,0.5}s\in\{0.05,0.5\}, 100 observed sizes per series · Simulation: E21 (R9, R10, R11), artefacts A3, A6 · Uses: C18

100C27 · descriptive · novelty: new, by the authors

For ETS(M,N,N) in the conventional multiplicative ETS form with Gamma errors, s=0.32s=0.32 and α=0.21\alpha=0.21, the approximate upper bound of the 90% prediction interval divided by the point forecast was 2.431 against 2.393 from simulated paths twelve steps ahead, and 2.616 against 2.536 after six unobserved periods at the end of the sample.

Scope: population: ETS(M,N,N) in the conventional form; conditions: Gamma errors, s=0.32s=0.32, α=0.21\alpha=0.21, 100 observations · Simulation: E22 (R5, R6, R7, R8), artefacts A3, A7

60C28 · descriptive · novelty: new, by the authors

For ADAM ETS(A,N,A) with lags 24 and 168 on 1344 hourly observations following a fixed sine pattern, the low-rank factor form of the SSOE Kalman filter was 2.96 times faster than the full covariance form for one missing value and 0.72 times as fast for 48, and slower than the full form above about 30 missing values.

Scope: population: ADAM ETS(A,N,A) with double seasonality; data: hourly time series with missing values; conditions: lags 24 and 168, 1344 observations, 1 to 48 missing values, data following a fixed sine pattern, one filter pass, one thread · Data: E23 (R4, R5), artefacts A3, A8 · Uses: C21 · Limitations: L2

62C29 · descriptive · novelty: new, by the authors

At variance ratio ft=4f_t=4, the 16-node Gauss-Hermite mixture of the error density over normal state uncertainty had a Kullback-Leibler divergence from the exact density of 0.0081 for the Laplace and 0.113 for the S distribution, against 0.0491 and 0.260 for the error kept in its family with scale sft1/psf_t^{1/p}.

Scope: population: pure additive models with Laplace or S errors; conditions: ft=4f_t=4, s=1s=1 · Simulation: E24 (R1, R2, R3, R4), artefacts A3, A9 · Uses: C24

Evidence

E1 · formal · cited by C1

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms · Assumes: C31, C32

  1. P(yt=0∣Yt−1)=(1−pt)∫f(z∣Yt−1)dz=1−ptP(y_t=0\mid Y_{t-1})=(1-p_t)\int f(z\mid Y_{t-1})dz=1-p_t (independence and continuity; from C31, C32)
  2. yt>0y_t>0 has density ptf(yt∣Yt−1)p_tf(y_t\mid Y_{t-1}) (independence; from C31)
  3. The joint density of the observed data is the product of these conditional terms (prediction-error decomposition; from E1.1, E1.2)

E2 · formal · cited by C2

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms · Assumes: C31, C32

  1. Q(θ∣θ(k))Q(\theta\mid\theta^{(k)}) contains at each zero the cross-entropy \int f_z^{(k)}\log f_z\,dz=-H_z(\theta^{(k)})-\KL(f_z^{(k)}\|f_z) (definition of the E-step; from D4)
  2. ∂θ∫fzlog⁡fzθdz|θ=θ(k)=∂θ∫fzdz=0\partial_\theta\int f_z\log f_z^{\theta}dz\Big|_{\theta=\theta^{(k)}}=\partial_\theta\int f_z\,dz=0 (differentiation under the integral; from E2.1)
  3. ∇Q(θ∣θ(k))|θ(k)=∇ℓ(θ(k))\nabla Q(\theta\mid\theta^{(k)})|_{\theta^{(k)}}=\nabla\ell(\theta^{(k)}), so EM fixed points are stationary points of ℓ\ell (EM identity; from E2.2, B1, C1)
  4. −∂θHz(θ)≠0-\partial_\theta H_z(\theta)\neq0 in general (e.g. in the scale), adding to the score of ℓ\ell (differentiation; from D4)

E3 · formal · cited by C3

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms

  1. Hz=12log⁡(2πeσ2)H_z=\frac12\log(2\pi e\sigma^2) for the normal distribution (definition; from D4)
  2. ∂ℓH∂σ2=−T12σ2+∑et22σ4−T02σ2,\frac{\partial\ell_H}{\partial\sigma^2}=-\frac{T_1}{2\sigma^2}+\frac{\sum e_t^2}{2\sigma^4}-\frac{T_0}{2\sigma^2}, zero at σ2=∑et2/T\sigma^2=\sum e_t^2/T (differentiation; from E3.1)
  3. EM iterates σ(k+1)2=(∑et2+T0σ(k)2)/T\sigma^2_{(k+1)}=(\sum e_t^2+T_0\sigma^2_{(k)})/T with fixed point 1T1∑et2\frac{1}{T_1}\sum e_t^2 (fixed point; from B1)

E4 · formal · cited by C4

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms · Assumes: C30, C33, C34, C35

  1. (𝐱t−1,zt,𝐱t)(\mathbf x_{t-1},z_t,\mathbf x_t) is jointly Gaussian given Yt−1Y_{t-1} (linearity of Gaussian models; from C35, C30)
  2. E⁡(zt∣Yt−1)=𝐰′𝐚tVar⁡(zt∣Yt−1)=σ2ftCov⁡(𝐱t,zt∣Yt−1)=σ2(𝐅𝐏t𝐰+𝐠)\begin{aligned} \E(z_t\mid Y_{t-1}) &= \mathbf w'\mathbf a_t \\ \Var(z_t\mid Y_{t-1}) &= \sigma^2f_t \\ \Cov(\mathbf x_t,z_t\mid Y_{t-1}) &= \sigma^2(\mathbf F\mathbf P_t\mathbf w+\mathbf g) \end{aligned} (moments of linear combinations; from E4.1)
  3. Conditioning on ztz_t gives the update; unobserved periods only propagate (Gaussian conditioning; from E4.2, C33)
  4. The likelihood follows from the observed-data decomposition, and σ2\sigma^2 concentrates out because ftf_t, 𝐤t\mathbf k_t do not depend on it (prediction-error decomposition; from E4.3, C1)

E5 · formal · cited by C15

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms

  1. With 𝐏t=𝟎\mathbf P_t=\mathbf 0, ft=1f_t=1 and 𝐤t=𝐠\mathbf k_t=\mathbf g (substitution; from D8)
  2. 𝐏t+1=𝐠𝐠′−𝐠𝐠′=𝟎\mathbf P_{t+1}=\mathbf g\mathbf g'-\mathbf g\mathbf g'=\mathbf 0 (substitution; from E5.1)
  3. By induction from 𝐏1=𝟎\mathbf P_1=\mathbf 0 the filter is the SSOE recursion (induction; from E5.2)

E6 · formal · cited by C5

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms

  1. After j−1j-1 unobserved periods from 𝐏=𝟎\mathbf P=\mathbf 0, f=1+∑i<jci2f=1+\sum_{i<j}c_i^2 with ci=𝐰′𝐅i−1𝐠c_i=\mathbf w'\mathbf F^{i-1}\mathbf g (iteration of the propagation; from D8)
  2. ft>1f_t>1 when c1≠0c_1\neq0, and 𝐤t≠𝐠\mathbf k_t\neq\mathbf g while 𝐏t≠𝟎\mathbf P_t\neq\mathbf 0 (substitution; from E6.1)
  3. The zero-error recursion uses variance σ2\sigma^2 and gain 𝐠\mathbf g, so its objective differs from the exact likelihood (comparison; from E6.2, D5, C4)

E7 · formal · cited by C16

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms

  1. P=(j−1)α2P=(j-1)\alpha^2, f=(j−1)α2+1f=(j-1)\alpha^2+1 after j−1j-1 unobserved periods (substitution $w=F=1$, $g=\alpha$; from D8)
  2. k=(P+α)/fk=(P+\alpha)/f (substitution; from E7.1, D8)

E8 · formal · cited by C12

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms

  1. log⁡zt=log⁡lt−1+ϕlog⁡bt−1+log⁡st−m+ut\log z_t=\log l_{t-1}+\phi\log b_{t-1}+\log s_{t-m}+u_t (logarithm; from D6)
  2. log⁡lt=log⁡lt−1+ϕlog⁡bt−1+αut\log l_t=\log l_{t-1}+\phi\log b_{t-1}+\alpha u_t, log⁡bt=ϕlog⁡bt−1+βut\log b_t=\phi\log b_{t-1}+\beta u_t in the ADAM form (logarithm; from D6)
  3. log⁡lt=log⁡lt−1+ϕlog⁡bt−1+log⁡(1+αϵt)\log l_t=\log l_{t-1}+\phi\log b_{t-1}+\log(1+\alpha\epsilon_t) in the conventional form (logarithm; from D7)

E9 · formal · cited by C6

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms · Assumes: C30, C33, C34, C36

  1. log⁡zt\log z_t follows a linear Gaussian SSOE model with known error mean −s/2-s/2 (linearity in logarithms; from C12, C36)
  2. The filter of the linear Gaussian model with the Jacobian −∑log⁡zt-\sum\log z_t gives the exact likelihood (change of variables; from E9.1, C4)
  3. Predictions are linear in −s/2-s/2 with gain-only coefficients, so vt(s)=vt0+sbtv_t(s)=v_t^0+sb_t (linearity; from E9.2)
  4. The score in ss times 2s22s^2 gives (∑bt2/ft)s2+T1s−∑(vt0)2/ft=0(\sum b_t^2/f_t)s^2+T_1s-\sum(v_t^0)^2/f_t=0 (differentiation; from E9.3)

E10 · formal · cited by C7, C8

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms · Assumes: C36

  1. ft=1f_t=1, bt=1/2b_t=1/2 give s2+4s−4m=0s^2+4s-4m=0 (substitution; from C6, C15)
  2. The positive root is 2(1+m−1)2(\sqrt{1+m}-1); 2(1−1−m)2(1-\sqrt{1-m}) solves s2−4s+4m=0s^2-4s+4m=0 instead (quadratic formula; from E10.1)
  3. At s=0.5s=0.5, E⁡m=s+s2/4=0.5625\E m=s+s^2/4=0.5625 and 2(1−1−0.5625)=0.6772(1-\sqrt{1-0.5625})=0.677 (arithmetic; from E10.2, C36)

E11 · formal · cited by C17

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms

  1. et2=(1+et)2−2(1+et)+1e_t^2=(1+e_t)^2-2(1+e_t)+1 and et2/(1+et)=(1+et)−2+(1+et)−1e_t^2/(1+e_t)=(1+e_t)-2+(1+e_t)^{-1} (algebra)
  2. Independence of ϵt\epsilon_t and DtD_t with E⁡(1+ϵ)2=1+s\E(1+\epsilon)^2=1+s and, for the inverse Gaussian, E⁡(1+ϵ)−1=1+s\E(1+\epsilon)^{-1}=1+s (independence and moments; from E11.1)

E12 · formal · cited by C10

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms · Assumes: C31, C32

  1. P(yt∈[0,u])=(1−pt)+ptFz(u)P(y_t\in[0,u])=(1-p_t)+p_tF_z(u) when zeros count as covered (law of total probability; from C31, C32)
  2. Setting it to cc gives FzF_z coverage (c−(1−pt))/pt(c-(1-p_t))/p_t, a quantile of the sizes around μ^t\hat\mu_t (algebra; from E12.1)

E13 · formal · cited by C19

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms · Assumes: C30

  1. x↦xgx\mapsto x^{g} is concave for 0<g<10<g<1, so E⁡(1+ϵ)g<(E⁡(1+ϵ))g=1\E(1+\epsilon)^{g}<(\E(1+\epsilon))^{g}=1 (Jensen's inequality; from D6)
  2. For ETS(M,N,N), E⁡(zT+h∣YT)=μ^T+h[E⁡(1+ϵ)α]h−1\E(z_{T+h}\mid Y_T)=\hat\mu_{T+h}[\E(1+\epsilon)^{\alpha}]^{h-1} by independence of the errors (independence; from E13.1, C30)
  3. For Gamma with mean one and variance ss, E⁡(1+ϵ)α=Γ(1/s+α)sα/Γ(1/s)\E(1+\epsilon)^{\alpha}=\Gamma(1/s+\alpha)s^{\alpha}/\Gamma(1/s) (Gamma moments; from E13.2)

E14 · formal · cited by C21

Analytical derivation in the source draft (artefact A1).

Foundation: Kolmogorov probability axioms

  1. An unobserved period adds the rank-one term 𝐠𝐠′\mathbf g\mathbf g' (propagation; from D8)
  2. An observed period removes the direction it reveals while adding the new error, so the rank does not increase (update; from D8)
  3. With 𝐌=[𝐔tF,𝐠]\mathbf M=[\mathbf U_t^F,\mathbf g] and 𝐌(𝐚,1)′=ft𝐤t\mathbf M(\mathbf a,1)'=f_t\mathbf k_t, a Householder reflection gives 𝐔t+1=𝐔tF−(ft𝐤t+ft𝐠)𝐚′ft+ft\mathbf U_{t+1}=\mathbf U_t^F-\frac{(f_t\mathbf k_t+\sqrt{f_t}\mathbf g)\mathbf a'}{f_t+\sqrt{f_t}} (Householder reflection; from E14.2)

E20 · simulation · primary · cited by C9, C25

Comparison, for iETS(M,N,N) under the ADAM and conventional multiplicative ETS forms, of the normal-drift mixture density (drift moments by 32-node Gauss-Legendre, mixture by 16-node Gauss-Hermite) and of the one-step density against the exact density of the size after an interval, computed on a grid of step 0.001 in log z by FFT convolution with tails cut at 1e-12, by the largest Kullback-Leibler divergence per observation over ss from 0.05 to 1 in steps of 0.05, α\alpha from 0.1 to 0.6 in steps of 0.05 and jj from 2 to 20; a 164-cell refinement run halved the step and cut tails at 1e-14. The run recorded smooth commit f9d062ee with the claimset scripts not yet committed; the scripts are archived at commit f1de3ea1.

Setup: studied normal-drift mixture density, one-step density · cases iETS(M,N,N), ADAM multiplicative ETS, conventional multiplicative ETS · procedure exact density by convolution · lens Kullback-Leibler divergence

E20.T1 · Largest Kullback-Leibler divergence per observation from the exact density of the size after an interval, of the normal-drift mixture density and of the one-step density, for iETS(M,N,N) with Gamma, log-normal and inverse Gaussian sizes in the conventional and the ADAM form (model: iETS(M,N,N))
Rowdistributionets_formlargest KL divergence per observation, normal-drift mixture vs exact densitylargest KL divergence per observation, one-step vs exact density
r1dgammaconventional0.00112.3593
r2dlnormconventional0.00455.9473
r3dinvgaussconventional0.00447.5452
r4dgammaadam0.01454.5836
r5dlnormadam0.000018.6354
r6dinvgaussadam0.0004630.772

E21 · simulation · primary · cited by C18, C26

Monte Carlo study of the scale of Gamma and inverse Gaussian iETS(M,N,N) sizes under the ADAM and conventional forms. Part A: 480 cells of simulated relative errors with drift (s∈{0.05,0.2,0.5,1}s\in\{0.05,0.2,0.5,1\}, α∈{0.1,0.3,0.6}\alpha\in\{0.1,0.3,0.6\}, intervals fixed at 2, 5 or 10 or geometric with occurrence probability 0.5 or 0.2, 100 or 1000 observed sizes), 2000 replications per cell, fixed-point iteration with tolerance 1e-10 and at most 100 iterations; a fit fails when an iterate is not positive and finite, the iteration does not converge, or the value leaves the admissible range. Part B: 120 cells within fitted models with s∈{0.05,0.5}s\in\{0.05,0.5\} and 100 observed sizes. Cells are grouped by the drift order E⁡(j−1)α2s\E(j-1)\alpha^2s. The run recorded smooth commit f9d062ee with the claimset scripts not yet committed; the scripts are archived at commit f1de3ea1.

Setup: studied fixed-point moment estimator, zero-error recursion · cases iETS(M,N,N), ADAM multiplicative ETS, conventional multiplicative ETS · procedure Monte Carlo study of the scale · lens relative bias, failed fits

  • R9 cells where the gain update raised the scale estimate by more than three Monte Carlo standard errors = 97 (estimator: moment equation solved by bisection, observed_sizes: 100, comparison: fitted model with gain update vs simulated drift errors)
  • R10 number of cells = 120 (estimator: moment equation solved by bisection, observed_sizes: 100, comparison: fitted model with gain update vs simulated drift errors)
  • R11 median increase in relative bias from the gain update = 0.1279 (estimator: moment equation solved by bisection, observed_sizes: 100, comparison: fitted model with gain update vs simulated drift errors)
E21.T1 · Median relative bias and share of failed fits of the fixed-point estimator of the Gamma and inverse Gaussian scale on 1000 observed sizes simulated with the drift, by drift order (estimator: fixed point, observed_sizes: 1000)
Rowdrift_ordermedian relative bias of the scale estimateshare of failed fits
r1<=0.020.00020
r20.02-0.1-0.00060.1767
r30.1-0.3-0.00420.5234
r4>0.3-0.00430.7101

E22 · simulation · primary · cited by C20, C27

Comparison of the approximate upper bound of the 90% prediction interval (0.95 quantile) of ETS(M,N,N) with s=0.32s=0.32, α=0.21\alpha=0.21, 100 observations and initial level 10, divided by the point forecast, with the quantile of 1,000,000 simulated paths, twelve steps ahead and 18 steps from the last observation after six unobserved periods at the end of the sample, for Gamma, inverse Gaussian and log-normal errors under the ADAM and conventional forms. The run recorded smooth commit 3eb93bb1 with the claimset scripts not yet committed; the scripts are archived at commit f1de3ea1.

Setup: studied approximate prediction interval · cases ETS(M,N,N) with s = 0.32 and alpha = 0.21, ADAM multiplicative ETS, conventional multiplicative ETS · procedure simulated paths · lens upper bound over the point forecast

  • R1 approximate upper bound of the 90% interval over the point forecast = 2.4309 (distribution: Gamma, ets_form: adam, h: 12, unobserved_at_end: 0, s: 0.32, alpha: 0.21)
  • R2 simulated upper bound of the 90% interval over the point forecast = 1.8071 (distribution: Gamma, ets_form: adam, h: 12, unobserved_at_end: 0, s: 0.32, alpha: 0.21)
  • R3 approximate upper bound of the 90% interval over the point forecast = 2.6159 (distribution: Gamma, ets_form: adam, h: 12, unobserved_at_end: 6, s: 0.32, alpha: 0.21)
  • R4 simulated upper bound of the 90% interval over the point forecast = 1.6379 (distribution: Gamma, ets_form: adam, h: 12, unobserved_at_end: 6, s: 0.32, alpha: 0.21)
  • R5 approximate upper bound of the 90% interval over the point forecast = 2.4309 (distribution: Gamma, ets_form: conventional, h: 12, unobserved_at_end: 0, s: 0.32, alpha: 0.21)
  • R6 simulated upper bound of the 90% interval over the point forecast = 2.3927 (distribution: Gamma, ets_form: conventional, h: 12, unobserved_at_end: 0, s: 0.32, alpha: 0.21)
  • R7 approximate upper bound of the 90% interval over the point forecast = 2.6159 (distribution: Gamma, ets_form: conventional, h: 12, unobserved_at_end: 6, s: 0.32, alpha: 0.21)
  • R8 simulated upper bound of the 90% interval over the point forecast = 2.5365 (distribution: Gamma, ets_form: conventional, h: 12, unobserved_at_end: 6, s: 0.32, alpha: 0.21)

E23 · quant · benchmark · primary · n 25 · cited by C22, C28

Timing of the SSOE Kalman filter in the smooth package with the low-rank factor against the full covariance, both built from smooth commit 53104ee9, for ADAM ETS(A,N,A) with lags 24 and 168 (193 profile cells), normal distribution and backcast initial states, on 1344 hourly observations with 1 to 48 nested missing values, for data generated by ETS(A,N,A) and by a fixed sine pattern; 25 interleaved repetitions per case on one cloud VM (Intel Xeon 2.80 GHz, 4 vCPUs, Linux 6.18.44, Python 3.11.15) with one thread.

Setup: studied low-rank factor form, full covariance form · cases hourly series with missing values · procedure interleaved timing · lens speed-up

  • R1 speed-up of low-rank over full covariance filter = 10.5511 ratio (data: ets, missing_values: 1, timed_call: filter pass)
  • R2 speed-up of low-rank over full covariance filter = 3.5046 ratio (data: ets, missing_values: 48, timed_call: filter pass)
  • R3 largest absolute difference in log-likelihood, low-rank vs full = 0.0000 (data: ets)
  • R4 speed-up of low-rank over full covariance filter = 2.9609 ratio (data: sine, missing_values: 1, timed_call: filter pass)
  • R5 speed-up of low-rank over full covariance filter = 0.7235 ratio (data: sine, missing_values: 48, timed_call: filter pass)
  • R6 largest absolute difference in log-likelihood, low-rank vs full = 0 (data: sine)

E24 · simulation · primary · cited by C24, C29

Kullback-Leibler divergence, in both directions, between the exact convolution of Laplace, S and Generalised Normal (shape 0.5 to 3) errors with normal state uncertainty and three approximations (the error kept in its family with scale sft1/psf_t^{1/p}, the one-step density, and Gauss-Hermite mixtures with 8 to 64 nodes), for 22 values of ftf_t from 1.01 to 100, by composite 8-node Gauss-Legendre integration. The run recorded smooth commit 3eb93bb1 with the claimset scripts not yet committed; the scripts are archived at commit f1de3ea1.

Setup: studied family-scaled error, Gauss-Hermite mixture · cases additive errors with state uncertainty · procedure divergence by numerical integration · lens Kullback-Leibler divergence

  • R1 KL divergence, exact density vs approximation = 0.0491 (distribution: dlaplace, approximation: family_scaled, f_t: 4, s: 1)
  • R2 KL divergence, exact density vs approximation = 0.2598 (distribution: ds, approximation: family_scaled, f_t: 4, s: 1)
  • R3 KL divergence, exact density vs approximation = 0.0081 (distribution: dlaplace, approximation: gauss_hermite 16 nodes, f_t: 4, s: 1)
  • R4 KL divergence, exact density vs approximation = 0.1127 (distribution: ds, approximation: gauss_hermite 16 nodes, f_t: 4, s: 1)

W1 · argument · deductive · premises C2

With a dynamic scale model, the scale at each period with zero demand is a free function of the parameters and enters the entropy-augmented likelihood only through the term −Hz-H_z; for the normal distribution −Hz=−12log⁡(2πeσ2)-H_z=-\frac12\log(2\pi e\sigma^2) increases without bound as σ→0\sigma\to0, and for the Gamma distribution the entropy decreases in the scale for large values, so the objective has no finite maximum.

W2 · argument · deductive

The expected log-density of an unobserved size is ∫fzlog⁡fzdz\int f_z\log f_z\,dz, which equals −Hz-H_z when the differential entropy is Hz=−E⁡log⁡fzH_z=-\E\log f_z; the term added for each zero therefore enters the objective with a negative sign.

W3 · argument · deductive · premises C15

The SSOE Kalman filter coincides with the single source of error recursion whenever its covariance is zero; the covariance starts at zero after backcasting and becomes non-zero only at an unobserved period, so on series without unobserved periods the last forward pass performs the same operations as the recursion.

Limitations

L1 · model specification · could bias magnitude · affects C18 · mitigation: none

Within a model fitted by the zero-error recursion, the fixed-point moment estimate of the Gamma and inverse Gaussian scale is biased upwards, because the states after an interval are updated with gain 𝐠\mathbf g and part of the drift leaks into the following errors.

L2 · computational · reduces transferability · affects C22, C28 · mitigation: none

The low-rank timing comparison uses one model configuration and one series length on one cloud VM (Intel Xeon 2.80 GHz, 4 vCPUs).

References

  1. Dempster et al. (1977) Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977). Maximum Likelihood from Incomplete Data Via the EM Algorithm. Journal of the Royal Statistical Society Series B: Statistical Methodology, 39(1), 1–22. doi:10.1111/j.2517-6161.1977.tb01600.x Used by: C2
  2. Ord et al. (1997) Ord, J. K., Koehler, A. B., & Snyder, R. D. (1997). Estimation and Prediction for a Class of Dynamic Nonlinear Statistical Models. Journal of the American Statistical Association, 92(440), 1621–1629. doi:10.1080/01621459.1997.10473684 Used by: C15
  3. Snyder (1985) Snyder, R. D. (1985). Recursive Estimation of Dynamic Linear Models. Journal of the Royal Statistical Society Series B: Statistical Methodology, 47(2), 272–276. doi:10.1111/j.2517-6161.1985.tb01355.x Used by: C15
  4. Svetunkov and Boylan (2023) Svetunkov, I., & Boylan, J. E. (2023). iETS: State space model for intermittent demand forecasting. International Journal of Production Economics, 265, 109013. doi:10.1016/j.ijpe.2023.109013 Used by: C14, D4, D5

Artefacts

Score

Reliability of the paper: 61, Medium

Scores are computed by rules from the claimset file (EntryLib rubric 0.2): they check the record, not the science, and may change as the rubric and reviews develop. The rubric. Every issue as text, for an agent preparing the next version.

The paper scores as its lowest principal claim, C6. Open problems in the principal claims and what they rest on, each counted once, with the most it costs a claim:

  1. −30 Support Rests on an unchecked proof. In C1, C2, C5, C6, C8, C10, C12. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. In C1, C3, C6, C7, C9, C10, C11. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. In C1, C4, C5, C6, C7. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −3 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. In C1, C2, C10. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. In C1, C2, C10. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. In C4, C5, C6. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. In C4, C5, C6. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. In C4, C5. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  9. −3 Open problems Assumption C36 is not checked: In the log-normal ADAM multiplicative ETS, $\log(1+\epsilon_t)\sim\mathcal N(-s/2,s)$, so that $\E(1+\epsilon_t)=1$. In C6, C7, C8. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

Claims

C1: 65, Medium

Novelty: new, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C2: 65, Medium

Novelty: refutes earlier work, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C3: 65, Medium

Novelty: new, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C4: 62, Medium

Novelty: extends earlier work, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C5: 62, Medium

Novelty: refutes earlier work, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C6: 61, Medium

Novelty: new, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  9. −1 Open problems Assumption C36 is not checked: In the log-normal ADAM multiplicative ETS, $\log(1+\epsilon_t)\sim\mathcal N(-s/2,s)$, so that $\E(1+\epsilon_t)=1$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C7: 61, Medium

Novelty: new, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  9. −1 Open problems Assumption C36 is not checked: In the log-normal ADAM multiplicative ETS, $\log(1+\epsilon_t)\sim\mathcal N(-s/2,s)$, so that $\E(1+\epsilon_t)=1$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C8: 61, Medium

Novelty: refutes earlier work, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C36 is not checked: In the log-normal ADAM multiplicative ETS, $\log(1+\epsilon_t)\sim\mathcal N(-s/2,s)$, so that $\E(1+\epsilon_t)=1$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  9. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C9: 95, High

Novelty: new, by the authors · no close match on EntryLib

  1. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.

C10: 66, Medium

Novelty: new, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C11: 61, Medium

Novelty: new, by the authors · no close match on EntryLib

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  9. −1 Open problems Assumption C36 is not checked: In the log-normal ADAM multiplicative ETS, $\log(1+\epsilon_t)\sim\mathcal N(-s/2,s)$, so that $\E(1+\epsilon_t)=1$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C12: 75, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.

C13: 60, Medium

Novelty: new, by the authors

  1. −30 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C14: 80, High

Novelty: refutes earlier work, by the authors

  1. −20 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.

C15: 62, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C16: 62, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C17: 75, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.

C18: 63, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −12 Open problems Limitation L1 (could bias magnitude, not mitigated): Within a model fitted by the zero-error recursion, the fixed-point moment estimate of the Gamma and inverse Gaussian scale is biased upwards, because the states after an interval are updated with gain $\mathbf g$ and part of the drift leaks into the following errors. To remove it: Mitigate it with evidence, or narrow the claim's scope so that it no longer applies.

C19: 72, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C20: 72, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C21: 62, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C22: 60, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −2 Open problems Limitation L2 (reduces transferability, not mitigated): The low-rank timing comparison uses one model configuration and one series length on one cloud VM (Intel Xeon 2.80 GHz, 4 vCPUs). To remove it: Mitigate it with evidence, or narrow the claim's scope so that it no longer applies.
  5. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  9. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C23: 62, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C24: 62, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C25: 100, High

Novelty: new, by the authors

Nothing lost.

C26: 63, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −12 Open problems Limitation L1 (could bias magnitude, not mitigated): Within a model fitted by the zero-error recursion, the fixed-point moment estimate of the Gamma and inverse Gaussian scale is biased upwards, because the states after an interval are updated with gain $\mathbf g$ and part of the drift leaks into the following errors. To remove it: Mitigate it with evidence, or narrow the claim's scope so that it no longer applies.

C27: 100, High

Novelty: new, by the authors

Nothing lost.

C28: 60, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −2 Open problems Limitation L2 (reduces transferability, not mitigated): The low-rank timing comparison uses one model configuration and one series length on one cloud VM (Intel Xeon 2.80 GHz, 4 vCPUs). To remove it: Mitigate it with evidence, or narrow the claim's scope so that it no longer applies.
  5. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  9. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

C29: 62, Medium

Novelty: new, by the authors

  1. −25 Support Rests on an unchecked proof. To remove it: Support the claim with evidence the claimset records, or have its proof checked.
  2. −5 Context A principal claim related to no earlier work. To remove it: Relate the claim to the earlier work it extends, addresses or departs from.
  3. −3 Open problems Assumption C30 is not checked: The errors $\epsilon_t$ of the single source of error model are independent across periods. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  4. −1 Open problems Assumption C31 is not checked: In the intermittent demand model the occurrence $o_t$ is independent of the potential demand size $z_t$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  5. −1 Open problems Assumption C32 is not checked: The potential demand size $z_t$ has a continuous distribution, so that $P(z_t=0)=0$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  6. −1 Open problems Assumption C33 is not checked: Whether a period is observed carries no information about the unobserved value $z_t$ beyond the model, so that the missing values are missing at random. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  7. −1 Open problems Assumption C34 is not checked: The initial states of the single source of error model are fixed parameters, so that the state covariance is zero at the start of the sample. To remove it: Test the assumption, or analyse how sensitive the claim is to it.
  8. −1 Open problems Assumption C35 is not checked: The errors of the pure additive model are normally distributed with constant variance $\sigma^2$. To remove it: Test the assumption, or analyse how sensitive the claim is to it.

Versions

  • Version 2 · 11 November 2026 at 11:22 UTC · corrected
  • Version 1 (this page) · 10 November 2026 at 21:10 UTC · active

Every version is kept in the public registry (github.com/entrylib/registry).