In this paper
AbstractIntroductionMulti-Horizon ForecastingMulti-Horizon Forecast StrategiesTheoretical FoundationsParameter LearningMean Squared Error Decomposition AnalysisBias and Variance Analysis by SimulationControlled Basis Experiment.Classical Autoregressive DGPs.Interpolation and Extrapolation to Unseen HorizonsEmpirical EvaluationNeural Forecast ArchitectureBenchmark DatasetsMain Empirical ResultsInference Complexity and ParallelismDiscussion and Related WorkConclusionUniversal Operator Approximation TheoremContinuous Horizon Variance DominanceParameters Covariance CharacterizationEffective horizon informationMSE Decomposition Experiment DetailsBias and Variance Analysis by SimulationNLAR MSE DecompositionSTAR MSE DecompositionUnseen Horizons Experiment DetailsQualitative Forecast ExplorationHorizon-Basis AblationDataset DetailsTraining Methodology and HyperparametersNetwork HyperparametersParameter-Matched Strategy ComparisonRegret-Based Dataset Sampling for Cross-LearningThe Continuous-Horizon Learning ObjectiveAdditional Point Forecast MetricsUniversal Periodic Operator Approximation TheoremReferencesWorksMathematics
Continuous Horizon Forecasting
Abstract
Multi-horizon forecasting has been dominated by discrete forecast strategies that treat model outputs as fixed-vectors or recurrently create them. In this paper, we introduce the Continuous Horizon (CH) strategy that generalizes upon recursive, direct, and multi-step strategies by redefining forecasts as a continuous operator and the forecast horizon as a query variable rather than an index. We present the theoretical foundations for the CH strategy, a Universal Operator Approximation Theorem, and a continuous-horizon bias-variance decomposition that formally proves the estimation variance improvements of CH over the discrete strategies. We validate the CH strategy empirically through a simulation study, and on 16 large-scale datasets from the M-series benchmark, showing how CH achieves competitive RMSE accuracy, while uniquely enabling parallel inference, targeted horizon querying, and interpolation and extrapolation across unseen horizons.
Introduction
Multi-horizon forecasting has been dominated by discrete strategies: multi-step (MS; [7, 4]), direct (DIR; [15]) and recurrent (REC) [6]. Although tree-based machine learning models [35], neural architectures [49, 10, 44], and foundation models [20, 18, 2, 64, 26] have advanced rapidly, multi-horizon forecast strategies have remained fixed for nearly a decade: multi-step fixed-grid outputs or recurrent rollouts [54, 57], with their known trade-offs in error accumulation, output dimension, and inability to query arbitrary horizons [5].
| Property | Recursive | Direct | Multi-Step | Continuous Horizon |
| Low Bias | ✗ | ✓ | ✓ | ✓ |
| Low Variance | ✗ | ✗ | ✗ | ✓ |
| Inference parallelism | ✗ | ✓ | ✓ | ✓ |
| Targeted horizon querying | ✗ | ✓ | ✗ | ✓ |
| Unseen Horizon inter/extrapolation | ✗ | ✗ | ✗ | ✓ |
| Joint horizon modeling | ✗ | ✗ | ✓ | ✓ |
Table 1. Desirable properties of forecasting strategies.
We argue that this discrete view of multi-horizon strategies is a historical artifact rather than a necessity. Time series forecasting has long recognized future trajectories as inherently continuous phenomena [51, 33, 12, 34]. In this paper, we elevate this continuous data approach into a forecast strategy: Continuous Horizon (CH). We reveal that discrete strategies are coarse discretizations of CH’s continuous-horizon forecast operator, we unlock theoretical understanding and showcase the strategy’s empirical advantages:
Universal Approximation & Unseen Horizons: We theoretically and empirically show that the Continuous Horizon (CH) strategy can approximate any continuous function across arbitrary horizons, including fractional horizons and those unseen during training.
Continuous Bias-Variance Dominance: We extend the discrete MSE decomposition framework of Ben Taieb and Atiya [5] to continuous time, formally proving how CH leverages continuous effective horizon information to reduce estimation variance over traditional multi-step (MS) and direct (DIR) strategies.
Accuracy, Computational Efficiency & Parallelism: CH achieves competitive accuracy results on the M-series benchmark (16 large-scale datasets), while enabling flexible inference that supports both targeted single-horizon queries and batched parallel inference.
The remainder of this paper is structured as follows. § 2 introduces the multi-horizon forecasting notation. § 3 presents the theoretical foundations of the CH strategy. § 4 conducts simulation of the strategies’ bias-variance behavior. § 5 investigates the strategies’ horizon interpolation and extrapolation capabilities. § 6 presents the empirical results of the M-series and inference computational complexity analysis. § 7 hosts the discussion and § 8 concludes.
Figure 1. Multi-horizon forecast strategies and the continuous-horizon paradigm.
Multi-Horizon Forecasting
Here we introduce general multi-horizon forecasting task notation. Let the forecast creation date be denoted by the index . We denote the functional random variable as , we denote its historical discrete observations at FCD as , for a single horizon the variable of interest is the future behavior of , for horizons .
Following the functional time series view [33, 51], we assume that the target variable comes from a continuous process:
with a continuous data generating process (DGP), and is a zero-mean noise term. The DGP coincides with the conditional expectation, .
We denote the -step ahead forecast , where is the model. We denote vector outputs . The model’s input handling and the structure of its forecast output define the overarching multi-horizon forecasting strategy.
Multi-Horizon Forecast Strategies
Inspired by the taxonomy of Ben Taieb et al. [6], we distinguish the primary variants of the multi-horizon forecast strategy: recursive, direct, multi-step and continuous horizon.
Recursive (REC): A single one-step-ahead model, that is, iterated H-times, by feeding each prediction back as input for the next step. It is a simple and parameter efficient approach that suffers from error accumulation.
Direct (DIR): H separate models are trained, one per horizon. Each model directly predicts the value of step h without dependencies on intermediate predictions. It avoids error accumulation, but assumes conditional independence and is less statistically efficient.
Multi-Step (MS): Also known as multiple-input multiple output. A single model outputs all H-forecasts simultaneously as a vector. It preserves the stochastic dependencies between horizons, as it does not assume conditional independence.
Continuous Horizon (CH): A single model produces the forecast on horizon h through a continuous function conditioned on h.
The Continuous Horizon strategy generalizes the other three: recursive, direct, and multi-step forecasts are all discretizations of a continuous horizon operator, obtained by evaluating an operator at fixed integer grid points or by iterating a one-step map. Redefining the forecast from a discrete mapping to an operator that returns a forecast function, that can be thought as the asymptotic limit of a forecast horizon grid. In this work we use a flexible bilinear parametrization of CH, similar to that of the Hyndman and Shahid Ullah [33] and the branch-trunk architectures [39].
Theoretical Foundations
Parameter Learning
Estimation of a forecasting model is carried out according to
where the forecaster parameters are selected to minimize loss . Strategies can be identified by a way to select horizons on which to train. We can describe a common learning objective using a horizon-selector measure on the horizon interval that specifies the lead times entering the objective: a point mass for the REC and DIR strategies, the empirical measure for MS, and the uniform measure for the CH strategy.
Mean Squared Error Decomposition Analysis
A deep understanding of the performance of forecasting strategies can be gained through the analysis of the mean squared error decomposition into bias and variance components.
Lemma 1 (MSE decomposition). The forecast MSE at horizon can be decomposed as
This decomposition separates the error at each horizon into an irreducible noise term , a squared bias term , and an estimation variance term . The noise is the conditional variance of around the optimal forecast ; it is irreducible, set by the DGP rather than the model, and typically grows with the horizon. The squared bias measures how far the average fitted forecast sits from , and reflects the model’s lack
of flexibility to represent the target variable. The estimation variance measures the sensitivity of the fitted forecast to the particular training samples, and shrinks as the effective number of observations informing grows.
Lemma 2 (Bias decomposition). The squared bias decomposes as
The approximation term measures the discrepancy between the optimal forecast and the best forecast attainable by the model, independent of any estimation of a finite-sample; the estimation term measures how the training-average estimator departs from that best-in-class forecast under a finite sample.
Model selection trades bias against variance ; the two results below show how the Continuous Horizon strategy controls each in turn.
Theorem 1 (Universal Operator Approximation Theorem). Let be a compact input domain of historical observations and be a compact forecast horizon interval. Let the DGP process of the target variable be governed by a continuous operator such as , with . Then for any accuracy , there exists a neural network approximation , composed of a finite set of continuous neural basis functions and coefficient mappings , such that:
Theorem 1 resembles the operator approximation results of Chen and Chen [14], which also support the DeepONet [39] and Fourier neural operators [38], and the neural basis approximation theorem of NHITS [10]. We include Theorem 1 to show that the separable bilinear decoder in Eq. (5) loses no expressiveness and to bound the approximation term of (Lemma 2). Proof available in Appendix § 9.
Theorem 2 (Continuous Horizon Variance Dominance). Consider residuals and let be a stationary, non-increasing, strictly convex kernel that describes the residual covariance at horizons , so that . For the multi-step discrete forecast grid , we define the effective horizon information ratio as:
Then, the effective horizon information of the CH, MS, and DIR strategies obeys the inequality:
Consequently, for a matched capacity of shared parameters , the estimation variance satisfies the following:
The proof of the variance dominance theorem is available in Appendix B.
Bias and Variance Analysis by Simulation
To validate Theorem 2, we do two simulation studies.
Controlled Basis Experiment.
We replace the learned neural components with a fixed basis of polynomials and harmonics, and we use a Gaussian process as the DGP. Thus, the experiment measures only the estimation variance. Figure 2 shows that, with the matched parameters, CH is confirmed to have a lower estimation variance than MS and DIR on all horizons.
Figure 2. Effective information and estimation variance for different MS resolution and forecast horizon. The estimation variance of CH dominates both MS and DIR.
Classical Autoregressive DGPs.
The second study uses the DGPs and the Monte Carlo protocol of Ben Taieb and Atiya [5]. We use an MQForecaster [63] with a CH decoder. The model must learn both the coefficient mappings and the basis functions . We use three autoregressive processes: a linear autoregression (LAR; 9), a nonlinear autoregression (NLAR; 43, 37), and a smooth-transition autoregression (STAR; 19, 59). Appendix § 11 gives the details of the DGPs.
Figure 3(a) shows the MSE decomposition of the LAR process for different training lengths . On this linear process, MS, DIR, and CH have a bias near zero. DIR has a high estimation variance, because it fits independent parameters for each horizon. MS decreases the variance, because it shares parameters across a discrete grid of horizons. CH has the lowest estimation variance and the lowest total error at all horizons . Appendix § 11.§ 2 and Appendix § 11.§ 3 repeat the bias–variance analysis for NLAR and STAR.
Figure 3. (a) MSE decomposition for a LAR DGP. (b) Bias–variance components across noise levels for NLAR, STAR, and LAR processes. CH improves the bias–variance tradeoff compared to MS and DIR strategies, an advantage that grows in higher-noise regimes. The dashed Naive-1 reference ( row) bounds the useful noise range: the learned strategies stay below it where the process is predictable; its collapse toward the irreducible noise floor marks the loss of predictability.
Figure 3(b) shows the bias–variance components when we increase the observation noise, which is a proxy for a weaker correlation between horizons. When noise increases, the estimation variance of DIR and MS increases rapidly. The shared continuous basis of CH regularizes the fit across horizons. At the highest noise level, CH has approximately 40 to 55% lower variance than DIR on all three DGPs. CH also has the lowest total error on NLAR and STAR, but on LAR, REC has a lower total error.
Interpolation and Extrapolation to Unseen Horizons
Discrete strategies such as MS and DIR give forecasts only at fixed horizons. CH can query its basis functions at any real . Thus, CH can interpolate at fractional horizons, for example , and it can extrapolate to unseen horizons. Of the discrete strategies, only the recurrent rollout of REC can also extrapolate. To show these capabilities, we simulate continuous-time Gaussian processes in five persistence regimes: mean-reverting, aperiodic, periodic, and slow and fast quasi-periodic (Figure 4, Appendix § 12).
Figure 4. Top: A sampled history (gray) and the exact posterior mean with its 10–90% band. The dotted line shows the end of the training grid (). Bottom: The RMSE regret at each horizon (Eq. (14)). A regret of zero is optimal. CH has the lowest regret on almost all seen and unseen horizons. MS and DIR are shown only on their training grid.
We train the MQForecaster model on the integer horizons . We evaluate it on three sets of horizons: the seen integers, the unseen half-integers (interpolation), and the unseen integers up to (extrapolation). For the half-integer horizons, MS and DIR use linear interpolation between their two adjacent output horizons. We measure the accuracy with the RMSE and with the regret relative to the exact analytical posterior.
where is a set of uniformly spaced query horizons, and is the exact conditional mean of § 2. For the seen and the interpolation sets, . For the extrapolation set, .
Figure 4 (bottom) and Table 2 show that CH has the lowest error on all five processes and on all three sets of horizons. Under interpolation, CH has a lower error than the interpolated MS and DIR forecasts. Appendix § 13 explains this result as manifestation of aliasing of a discrete horizon grid.
| Process | Seen () CH | MS | DIR | REC | Interpolation CH | MS | DIR | REC | Extrapolation CH | REC |
| Mean-reverting | 0.873 | 0.934 | 0.915 | 0.939 | 0.872 | 0.932 | 0.908 | 0.937 | 1.211 | 1.280 |
| Aperiodic | 0.733 | 0.759 | 0.754 | 0.763 | 0.730 | 0.753 | 0.746 | 0.758 | 1.054 | 1.110 |
| Periodic | 0.568 | 0.581 | 0.581 | 0.617 | 0.568 | 0.576 | 0.575 | 0.613 | 0.607 | 0.649 |
| QP (slow) | 0.623 | 0.640 | 0.639 | 0.671 | 0.623 | 0.636 | 0.633 | 0.670 | 0.756 | 0.780 |
| QP (fast) | 0.689 | 0.716 | 0.712 | 0.743 | 0.688 | 0.712 | 0.705 | 0.740 | 0.940 | 0.975 |
Table 2. Interpolation and extrapolation on five Gaussian-process regimes (RMSE, lower is better). All strategies train on the horizons . We evaluate on the seen integers, the unseen half-integers (interpolation), and the unseen integers up to (extrapolation). For the half-integers, MS and DIR use linear interpolation between their output horizons. MS and DIR have no output for , but CH and REC give forecasts for these horizons. The best value in each block is bold.
In extrapolation, CH reverts to the unconditional mean on the aperiodic processes and follows the cyclic dynamics with small ripples. The REC rollout has a higher error than CH on all five processes. For periodic targets, Theorem 3 (Appendix J) extends the approximation guarantee beyond the training horizons .
Empirical Evaluation
Neural Forecast Architecture
We evaluate all strategies within the unified MQForecaster architecture [63, 46] sequence-to-sequence model [16]. Its modular encoder-decoder design provides a natural testbed to isolate the impact of the forecasting strategy while holding architecture constant (Figure 5). Historical realizations are encoded into a hidden state .
Figure 5. Continuous Horizon (CH) decoder within the MQForecaster architecture. Instead of outputting a discrete vector, the CH forecast strategy outputs a function , capable of evaluating continuous horizons on demand. This enables direct evaluation at arbitrary continuous lead times (including unseen interpolated and extrapolated steps) while maintaining parameter sharing for lower estimation variance. Holding the backbone fixed isolates the strategy, confirming that the CH advantages generalize across encoders.
A decoder then maps the hidden state to the forecast. We keep the encoder fixed and change only the decoder to realize each forecast strategy.
The strategies MS and DIR instantiate the decoder exactly as their taxonomy definitions (§ 2) prescribe, replacing the abstract function with an MLP acting on the hidden state; therefore, we omit their repetition here. For the REC strategy, we employ a similar approach to that of the DeepAR decoding strategy [54].
To enable the CH strategy, we use a bilinear expansion approximator informed by the universal approximation theorem (Theorem 1):
where the constant basis function gives a level that is the same for all horizons. The horizon basis functions are a learned embedding table; see the ablation study in Appendix F, where we explore basis variants. Based on this ablation, we use tabular embedding basis functions (). Appendix H.§ 4 describes the controlled learning objective of the M-series experiments.
Benchmark Datasets
We evaluate our approach on the M-series benchmark, comprising 16 large-scale datasets with more than 100,000 series from the competitions M1 [40], M3 [41], M4 [42], and Tourism [3]. Following recent literature [47, 50], we use temporal cross-validation by holding out the final observations per series (Details in Appendix § 15): the first form the validation window, and the final comprise the test window (Figure 6). Rolling the test FCD forward yields overlapping evaluations covering . Denoting the final series index by , the data partitions are defined as:
Figure 6. Temporal cross-validation on an M4 hourly series, with training (blue), validation (yellow, ), and test (salmon, ) regions.
Main Empirical Results
Accuracy Metrics. Following Hyndman and Koehler [32], we report the Relative Average RMSE (relAvgRMSE), defined as the ratio of the mean per-series RMSE to that of the Naive baseline:
where is the number of series and , , and denote the target, model forecast and Naive baseline for series . Computing the root per series before aggregation prevents large-scale series from dominating the metric. Values below 1 indicate an improvement over Naive.
Forecast Baselines. We vary MQForecaster’s encoder (Section 6.1): CNN [48], LSTM [24, 11], a Transformer [62], and S4 [22]. We include ETS [25] and ARIMA [31, 29] as baselines. Appendix H.1 gives the network hyperparameters. Appendix H.2 discusses the parameter matching across strategies.
Point Forecast Results. Table 3 compares the strategies at matched parameter counts (Appendix H.2). Thus, a difference in accuracy comes from the forecast strategy, not from the capacity of the model. CH has a lower relAvgRMSE than MS in 40 of the 64 encoder and frequency pairs, with a mean improvement of 2.6%. In 12 pairs, the improvement is greater than two standard errors and in 8 pairs MS is better by more than two standard errors. The advantage of CH increases with the horizon length, as predicted by § 3. CH is better in all M4 Hourly pairs (), with a decrease of 38% (CNN) and 49% (Transformer), and in 14 of the 16 monthly pairs. In the short yearly horizons (), the two strategies are approximately equal, and on M4 Daily, MS is lower in all four pairs. CH is also better than REC in 58 of the 64 encoder and dataset pairs, with a mean improvement of 19%.
| Dataset | Freq | LSTM CH | LSTM MS | LSTM DIR | LSTM REC | CNN CH | CNN MS | CNN DIR | CNN REC | Transformer CH | Transformer MS | Transformer DIR | Transformer REC | S4 CH | S4 MS | S4 DIR | S4 REC | StatsForecast ETS | StatsForecast ARIMA |
| M1 | M | 0.8098 (0.0060) | 0.8784 (0.0061) | 0.9244 (0.0048) | 0.9281 (0.0149) | 0.8226 (0.0050) | 0.9023 (0.0078) | 0.9150 (0.0222) | 0.9928 (0.0417) | 0.9022 (0.0226) | 0.9373 (0.0095) | 0.9715 (0.0215) | 0.9741 (0.0082) | 0.8705 (0.0120) | 0.8781 (0.0078) | 0.9366 (0.0277) | 0.9891 (0.0159) | 0.7347 (-) | 0.7625 (-) |
| M1 | Q | 0.9305 (0.0264) | 0.9441 (0.0207) | 0.9616 (0.0314) | 0.9172 (0.0268) | 0.9185 (0.0053) | 0.9359 (0.0179) | 0.9969 (0.0244) | 1.2225 (0.1147) | 1.0004 (0.0809) | 1.0238 (0.0178) | 1.0329 (0.0515) | 0.9454 (0.0251) | 0.9160 (0.0214) | 0.9031 (0.0047) | 1.0435 (0.0234) | 0.9264 (0.0235) | 0.8998 (-) | 0.9071 (-) |
| M1 | Y | 0.8426 (0.0245) | 0.7564 (0.0207) | 0.7958 (0.0240) | 0.9285 (0.0805) | 0.8014 (0.0122) | 0.8323 (0.0289) | 0.7367 (0.0473) | 0.7786 (0.0353) | 0.6896 (0.0267) | 0.8307 (0.0770) | 0.8448 (0.0720) | 0.8032 (0.0530) | 0.7384 (0.0328) | 0.7950 (0.0281) | 0.7750 (0.0434) | 0.8660 (0.0559) | 0.7854 (-) | 0.7696 (-) |
| M3 | O | 0.2348 (0.0020) | 0.2429 (0.0214) | 0.2414 (0.0069) | 0.2651 (0.0159) | 0.1844 (0.0012) | 0.2192 (0.0033) | 0.2422 (0.0078) | 0.3834 (0.0712) | 0.2507 (0.0121) | 0.2798 (0.0099) | 0.4130 (0.1321) | 0.2930 (0.0233) | 0.1870 (0.0030) | 0.1979 (0.0040) | 0.2568 (0.0153) | 0.3134 (0.0200) | 0.7146 (-) | 0.7259 (-) |
| M3 | M | 0.6973 (0.0042) | 0.7134 (0.0071) | 0.7807 (0.0062) | 0.8662 (0.0261) | 0.7222 (0.0011) | 0.7445 (0.0028) | 0.7465 (0.0036) | 1.1237 (0.0447) | 0.7618 (0.0169) | 0.7904 (0.0098) | 0.8213 (0.0168) | 1.0170 (0.0674) | 0.7115 (0.0034) | 0.7353 (0.0059) | 0.7503 (0.0046) | 0.9843 (0.0255) | 0.8380 (-) | 0.8490 (-) |
| M3 | Q | 0.7962 (0.0095) | 0.8061 (0.0111) | 0.8685 (0.0113) | 0.9332 (0.0224) | 0.7849 (0.0074) | 0.8035 (0.0069) | 0.8541 (0.0126) | 1.0807 (0.0890) | 0.8780 (0.0481) | 0.9300 (0.0333) | 0.9650 (0.0137) | 0.9288 (0.0454) | 0.7561 (0.0111) | 0.7390 (0.0038) | 0.9114 (0.0040) | 0.9702 (0.0142) | 0.9206 (-) | 0.9279 (-) |
| M3 | Y | 0.8879 (0.0130) | 0.7854 (0.0104) | 0.8758 (0.0108) | 0.9264 (0.0458) | 0.8230 (0.0100) | 0.7939 (0.0084) | 0.7772 (0.0152) | 0.9710 (0.0678) | 0.8685 (0.0674) | 0.8596 (0.0428) | 0.8605 (0.0236) | 0.8324 (0.0236) | 0.7996 (0.0240) | 0.7793 (0.0133) | 0.8778 (0.0114) | 0.9198 (0.0113) | 0.9678 (-) | 1.0116 (-) |
| M4 | H | 0.1719 (0.0169) | 0.1810 (0.0160) | 0.3916 (0.0033) | 0.6788 (0.0468) | 0.2850 (0.0057) | 0.4610 (0.0105) | 0.4739 (0.0072) | 0.7736 (0.0536) | 0.2471 (0.0117) | 0.4877 (0.0522) | 0.6908 (0.0161) | 0.9976 (0.0237) | 0.2698 (0.0063) | 0.3016 (0.0066) | 0.2773 (0.0059) | 0.7000 (0.0122) | 0.4158 (-) | 0.1987 (-) |
| M4 | D | 0.8362 (0.0711) | 0.7923 (0.0712) | 0.8536 (0.0133) | 0.9311 (0.0347) | 0.8344 (0.0067) | 0.8293 (0.0057) | 0.8531 (0.0052) | 0.9065 (0.0053) | 0.8687 (0.0270) | 0.8401 (0.0183) | 0.8717 (0.0203) | 1.0212 (0.0880) | 0.8178 (0.0125) | 0.8068 (0.0068) | 0.8458 (0.0173) | 1.0401 (0.0491) | 0.9815 (-) | 0.9879 (-) |
| M4 | W | 0.7839 (0.0124) | 0.7240 (0.0120) | 0.8100 (0.0090) | 0.8047 (0.0016) | 0.7067 (0.0072) | 0.7894 (0.0046) | 0.7787 (0.0072) | 0.8640 (0.0604) | 0.8288 (0.0128) | 0.8333 (0.0363) | 0.8363 (0.0163) | 0.8882 (0.0233) | 0.8055 (0.0158) | 0.7943 (0.0049) | 0.8271 (0.0078) | 0.9093 (0.0399) | 0.9507 (-) | 0.9320 (-) |
| M4 | M | 0.7114 (0.0133) | 0.7307 (0.0057) | 0.7482 (0.0110) | 0.8337 (0.0219) | 0.7278 (0.0072) | 0.7305 (0.0053) | 0.7322 (0.0055) | 1.0405 (0.0541) | 0.7198 (0.0550) | 0.7490 (0.0522) | 0.7841 (0.0161) | 1.0314 (0.0237) | 0.7054 (0.0159) | 0.7343 (0.0043) | 0.7297 (0.0059) | 0.9806 (0.0122) | 0.8802 (-) | 0.8904 (-) |
| M4 | Q | 0.7426 (0.0073) | 0.7373 (0.0093) | 0.7904 (0.0081) | 0.8072 (0.0124) | 0.7349 (0.0065) | 0.7398 (0.0053) | 0.7520 (0.0144) | 0.8950 (0.0541) | 0.8274 (0.0550) | 0.8158 (0.0522) | 0.8448 (0.0161) | 0.9164 (0.0237) | 0.7580 (0.0159) | 0.7591 (0.0102) | 0.7961 (0.0059) | 0.8868 (0.0122) | 0.8714 (-) | 0.9143 (-) |
| M4 | Y | 0.8034 (0.0092) | 0.7940 (0.0066) | 0.8831 (0.0013) | 0.8642 (0.0467) | 0.7515 (0.0065) | 0.7230 (0.0054) | 0.7412 (0.0144) | 0.8591 (0.0541) | 0.7566 (0.0550) | 0.8044 (0.0522) | 0.7979 (0.0161) | 0.7406 (0.0237) | 0.7133 (0.0159) | 0.7720 (0.0102) | 0.8074 (0.0059) | 0.8417 (0.0122) | 0.8578 (-) | 0.9376 (-) |
| Tourism | M | 0.5856 (0.0355) | 0.6011 (0.0061) | 0.7637 (0.0274) | 0.8260 (0.0419) | 0.6533 (0.0074) | 0.6772 (0.0074) | 0.6764 (0.0039) | 1.1139 (0.0583) | 0.8286 (0.0071) | 0.8171 (0.0070) | 0.8317 (0.0116) | 1.0842 (0.1034) | 0.6790 (0.0063) | 0.6217 (0.0034) | 0.6433 (0.0075) | 1.2160 (0.0622) | 0.4476 (-) | 0.5026 (-) |
| Tourism | Q | 0.6702 (0.0241) | 0.6691 (0.0098) | 0.8084 (0.0164) | 0.8133 (0.0813) | 0.6304 (0.0056) | 0.6563 (0.0051) | 0.7421 (0.0145) | 0.9466 (0.0319) | 0.8464 (0.0848) | 0.8733 (0.0850) | 0.8508 (0.0850) | 0.8725 (0.0472) | 0.6389 (0.0127) | 0.6064 (0.0034) | 0.7636 (0.0176) | 0.9239 (0.0343) | 0.6395 (-) | 0.7091 (-) |
| Tourism | Y | 1.0035 (0.0133) | 0.9265 (0.0295) | 0.9803 (0.0168) | 0.9683 (0.0216) | 0.8991 (0.0171) | 0.9517 (0.0315) | 0.9778 (0.0219) | 0.9480 (0.0248) | 0.9138 (0.0277) | 0.9556 (0.0456) | 0.9518 (0.0181) | 0.9431 (0.0413) | 0.9562 (0.0359) | 0.9523 (0.0122) | 1.0954 (0.0356) | 1.0042 (0.0022) | 0.8924 (-) | 1.0213 (-) |
Table 3. Empirical evaluation of point forecasts. Relative Average RMSE (relAvgRMSE) with the Naive baseline, averaged over 5 runs. Lower measurements are preferred; values below 1 beat the Naive. The methods without standard deviation have deterministic solutions. For each encoder we compare four forecasting strategies: our Continuous Horizon (CH), multi-step (MS), direct (DIR), and recurrent (REC).
Appendix I complements the RMSE results with sMAPE and MASE, making our results closer to the M competitions.
Inference Complexity and Parallelism
Beyond accuracy, strategies differ in the decoder cost of one forecast on a queried horizon (Table 4, Figure 7). MS computes the full grid of horizons, so a query costs . REC rolls out sequential steps, so one query costs . DIR runs only the head of horizon , so one query costs , but DIR must train and store heads.
| REC | ||
| DIR | † | |
| MS | † | |
| CH |
Table 4. Decoder cost of one query
Figure 7. Targeted query decoder cost (schematic). MS forecasts all horizons. DIR runs horizon-’s model. REC rolls out sequential steps. CH needs one readout of the cached basis. For , MS and DIR need a new model (dashed lines and †).
The CH basis functions do not depend on the history, so we cache them before inference. Then a query on any horizon needs only one readout, at a cost of . For , CH uses the same model. When the complete trajectory is necessary, CH evaluates all horizons in parallel, as MS does.
Discussion and Related Work
CH extends the functional time series approach [51, 33] in three ways. It uses the function directly as an end-to-end forecast strategy, without an in-sample decomposition. It learns the full mapping with a neural network, without smoothing or functional principal components. It learns its basis with MLPs and embeddings, in place of a fixed basis. A future direction is to constrain the learned basis, for example, with orthonormality [51].
Other neural forecast architectures scratched the surface of the CH forecast strategy, examples include DAM [17] and NBEATS [49, 45] that uses harmonics and polynomial trends, NHITS [10] that use wavelets. In contrast to these approaches, we do not pre-conditioning the function family and rather learn it via embeddings, but more importantly we allow for variable length horizon in contrast to these architectures that have MS fixed outputs.
Neural DEs [13, 52, 36] learn vector fields that are integrated to obtain continuous-horizon trajectories. In contrast, CH retains a discrete-time encoder and uses a horizon-separable basis for direct, parallel evaluation without numerical integration. Inspired by ODEs and integrated time-series models (e.g., ARIMA), a promising direction is to apply CH to first differences and integrate predictions to recover forecast levels, potentially improving robustness to non-stationary trends.
Our simulations decompose only the point-forecast errors. A probabilistic extension can use the decomposition of Hersbach [23], to study the uncertainty, calibration, and reliability tradeoffs of different forecast strategies.
Ben Taieb and Atiya [5] proposed discrete hybrid strategies, such as DirRec, that combine the low variance of recursive forecasts with the low bias of direct forecasts. A hybrid of discrete strategies and CH is a further direction for very long horizons.
Current foundation forecasting models use a fixed output grid or a recursive rollout [20, 2, 18, 64, 26] and truncate a long output to get a shorter horizon. CH gives only the necessary horizons, so its integration with foundation models is a promising direction.
Conclusion
We introduced the Continuous Horizon (CH) forecast strategy, and conducted a comprehensive theoretical investigation of the properties of CH, showing that it provides universal approximation guaranties for forecasting operators. Furthermore, we proved that, under reasonable assumptions, the estimation variance of CH improves on that of discrete strategies because CH makes a more efficient use of the information from the horizon.
We validated these theoretical findings through simulation studies and bias-variance decompositions, demonstrating that CH achieves a bias profile comparable to DIR and MS with a lower estimation variance. Beyond fixed grid training, CH demonstrates strong accuracy in both interpolation and extrapolation regimes, forecasting horizons unseen during training.
CH achieves competitive accuracy. It has a lower relAvgRMSE than MS in 40 of the 64 encoder and frequency pairs of the M-series benchmark. Finally, we highlight the substantial computational efficiency of CH, particularly its ability to execute targeted horizon queries in constant time.
Universal Operator Approximation Theorem
In this appendix, we prove the universal approximation theorem introduced in Section 3. We use the Stone-Weierstrass theorem to prove the density of finite sum approximation functions, and complement with the universal approximation theorem applied to its components.
Lemma 3. Let be a compact domain and a compact horizon interval. Let given by be a continuous forecast function. Then, for any , there exists a finite integer and continuous functions and for such that:
Proof. The Cartesian product of two compact sets is compact in . Consider the set of finite separable functions:
We verify that forms a subalgebra of that satisfies the conditions of the Stone-Weierstrass Theorem.
Contains Constants: Taking and shows .
Closed under Addition and Multiplication: For any , and by expansion of finite sums.
Point Separation: For , if , choose (where coordinate ) and . If , choose and .
By the Stone-Weierstrass Theorem, is dense in equipped with the uniform norm. Therefore, for any and any , there exists some element in within the distance of , establishing the existence of finite and continuous functions .
:::
Lemma 4. Let be a finite set of continuous functions
where is compact. Then, for all , there exist neural networks
such that
Proof. This lemma is a direct consequence of the neural universal approximation theorem [27]. Since every is continuous in the compact domain , a non-polynomial activated feedforward neural network can approximate uniformly to arbitrary precision. As the number of functions is finite, the approximation can be constructed for every using the same target precision .
Lemma 5. Let be a finite set of continuous coefficient mappings
where is compact. Then, for all , there exist neural networks
such that
Proof. As in Lemma 4, this follows from the neural universal approximation theorem. Since every is continuous in the compact domain , there exists a neural network that approximates it uniformly with arbitrary precision. Since is finite, the approximation can be constructed simultaneously for all coefficient mappings.
Theorem 1. Let be a compact domain of historical observations and let be a compact forecast horizon interval. Let a continuous forecast operator be
where is equipped with the uniform norm. Then, for all , there exists a finite , a set of continuous neural network basis functions , and neural coefficient mappings , collected in the parameter vector , such that the neural forecaster approximates uniformly over the horizon,
Because has finite length, this uniform bound implies the squared-error bound
which is the form that controls the squared approximation bias of Lemma 1.
Proof. Since
is continuous under the uniform norm, the associated joint forecast function
is continuous. Using Lemma 3, for all , there exist finite and continuous functions and such that
Let
denote this finite separable approximation. We now approximate the two ingredients of . Using Lemma 4, for all , there exist basis functions such that
Using Lemma 5, for all , there exist coefficients such that
Define the neural forecast approximation as
where the single finite parameter vector concatenates the weights of the coefficient networks (each with weights ) and the basis networks (each with weights ). Since is finite and each network has finitely many weights, is finite-dimensional.
To conclude, we combine the finite separable approximation, the neural basis approximation, and the neural coefficient approximation. Using the triangular inequality,
Bounding each supremum through the triangle inequality over the sum,
Using continuity in compact domains, the finite collections and are uniformly bounded. Let
Then
Finally, the three approximation accuracies can be chosen
which gives the uniform bound
Since has finite length , the uniform bound controls both the error and the squared error,
for all . The squared-error bound is the form that controls the squared approximation bias of Lemma 1.
Therefore, the continuous operator can be arbitrarily approximated by a finite neural basis expansion whose basis functions depend on the forecast horizon and whose coefficients depend on the historical observations.
Continuous Horizon Variance Dominance
In this appendix, we prove the variance dominance theorem introduced in § 3, which bounds the estimation variance of Lemma 1 across forecast strategies. The argument proceeds in two stages. First, we characterize how the horizon-selector measure shapes the covariance of the fitted parameters: a first-order expansion of the empirical-risk minimizer yields a sandwich covariance , whose variability matrix aggregates the cross-horizon residual covariances selected by , and the delta method propagates the parameter covariance to the horizon- forecast covariance. Second, we turn the forecast covariance into a scalar ordering: under a separable score covariance with a correlation kernel , the variability collapses to a single effective horizon information , the number of independent horizons the forecast strategy’s objective supplies during estimation. Direct forecasting draws on one horizon, multi-step on a discrete grid, and Continuous Horizon on the whole interval, so their effective information obeys at any finite grid resolution. Because variance scales inversely with at matched shared-parameter dimension, this chain gives the variance ordering .
Parameters Covariance Characterization
The variance term of Lemma 1 is governed by the covariance of the fitted parameters, which the following lemma characterizes for REC, DIR, MS, and CH strategies at once. Each strategy is identified by the horizon-selector measure of Eq. (6): a point mass for the direct strategy, the empirical measure for multi-step, and the uniform measure for the Continuous Horizon strategy.
Lemma 6 (Parameters’ covariance under a horizon selector). Let minimize the -averaged empirical risk of Eq. (6) over the tuples of . Define the per-horizon score at the population optimum and its -aggregate,
The estimator solves the first-order condition, linearizing around ,
where is the score of tuple , the gradient of its per-tuple risk , and is the expected Hessian of the risk at the optimum. Assuming (a) is consistent for , (b) is invertible, and (c) the expansion remainder is asymptotically negligible, we solve the linear system
Its right-hand side is an average of mean-zero scores with covariance , , by bilinearity of the covariance operator. Propagating through [28, 60], the fitted parameters therefore have covariance
with variability matrix given by
Substituting each selector into Eq. (B.2) yields its variability matrix.
Proposition 1 (Per-strategy variability). With the horizon selectors of Eq. (6), the variability matrix takes the forms
The direct strategy estimates an independent parameter block for each horizon, so its selector retains only the diagonal term and no cross-horizon covariance enters; multi-step and the Continuous Horizon strategy pool a single across horizons, so their accumulate the off-diagonal score covariances over the discrete grid and the continuum, respectively.
The three forms include progressively more off-diagonal score covariance: none for direct, a discrete double sum over the grid points for multi-step, and the full double integral over the continuum for Continuous Horizon. The estimator covariance of Lemma 6, with the per-strategy of Proposition 1, propagates to the forecast through the delta method, giving the horizon- estimation variance of Lemma 1.
Corollary 1 (Forecast variance). Let be the forecast gradient at the optimum. Linearization of the forecast around ,
so that, substituting the covariance of Eq. (B.1),
The forecast variance thus inherits the sandwich structure, quadratic in the forecast gradient and scaled by the parameter covariance that the selector shapes through and .
Effective horizon information
To turn the variability matrix into a scalar measure of how much independent information the horizons provide, we impose a separable structure on the score covariance.
Assumption 1 (Separable score covariance). The horizon-score process has covariance
where is a fixed parameter-space matrix and is a correlation kernel: continuous, positive semidefinite, with and , decaying in with finite correlation length.
The nonnegative, decaying correlation captures that nearby forecast horizons provide overlapping rather than independent information. This is the score-space analogue of the effective-sample-size construction in spatial statistics [21, 61], and of its functional extension [1], where correlated observations carry less information than their nominal count.
Lemma 7 (Effective horizon information). Under Assumption 1, define the effective horizon information of a strategy with selector as
Then the variability matrix factors as , and the shared-parameter covariance of Lemma 6 becomes
Corollary 2 (Effective information per strategy). Substituting the selectors of Eq. (6) into Eq. (B.8),
These satisfy two limits. When the horizons are fully correlated (), every strategy collapses to : all horizons carry the information of one. When the multi-step horizons are independent (), , the nominal count underlying the residuals-per-parameter argument of Ben Taieb and Atiya [5]; for positively correlated horizons . The independent case is thus a special case of multi-step, not an assumption imposed on Continuous Horizon.
For a stationary kernel with lag , when is large relative to the correlation length the Continuous Horizon effective information admits the closed form
Information grows with the interval length divided by the correlation length; note that a denser horizon sampling does not drive the variance to zero while remains positively correlated.
Remark 1 (Residuals per parameter). Under the additional isotropy assumption that the horizon-averaged sensitivity factor scales with the number of shared parameters , the forecast variance takes the residuals-per-parameter form
The sharpest comparison is with multi-step, since both strategies pool a single and differ only in how the horizon objective is sampled: a discrete
grid of points for multi-step, the whole interval for Continuous Horizon. Their effective information is governed by the same kernel , so is the Riemann sum that integrates, and refining the grid past the correlation length adds only correlated scores, so saturates at the continuum ceiling rather than growing with the nominal count. This is the continuous form of the classical bias–variance trade-off: the pooling that lowers the variance is bounded by the horizon correlation structure, and Continuous Horizon realizes that bound without committing to a fixed output grid.
Theorem 2. (Continuous Horizon Variance Dominance).
Suppose, in addition to Assumption 1, that the stationary correlation kernel is nonincreasing and strictly convex in , and that the multi-step horizons form a regular grid on . Then the continuous horizon, multi-step, and direct forecast strategies’ effective information satisfy
with equality between multi-step and continuous horizon only in the fine-grid limit. Consequently, for a matched capacity of shared parameters (the shared trunk parameters, held fixed between multi-step and Continuous Horizon so that only the horizon sampling differs), the forecast estimation variance satisfies
so the continuous-horizon estimator attains strictly more variance reduction than any regular multi-step grid at any finite resolution.
Proof sketch. The direct strategy carries by Corollary 2, while any positively correlated grid of more than one horizon gives , establishing the strict left inequality. For the right inequality, by Eq. (B.8) is equivalent to the grid average of exceeding its continuum average. For a strictly convex, nonincreasing on a regular grid, the rectangle rule strictly underestimates the integral of over each cell, so the grid double-average strictly dominates the double integral; inverting reverses the inequality. As the grid spacing vanishes, the rectangle rule converges to the integral, so equality holds only in the fine-grid limit. The variance ordering of Eq. (B.14) then follows from Eq. (B.9): at the matched shared-parameter capacity , the forecast variance scales as , so the information ordering inverts into the stated variance ordering. □
MSE Decomposition Experiment Details
Bias and Variance Analysis by Simulation
The bias and variance study of § 4 uses the three autoregressive processes of Ben Taieb and Atiya [5]. Each process has i.i.d. Gaussian innovations .
Smooth Transition Autoregression (STAR). This process has two regimes. A logistic function of moves it smoothly between them [59]:
with and . Thus, the process is almost deterministic.
Nonlinear Autoregression (NLAR). This process is a neural network with two hidden units, fit to the annual sunspot series [59]. It has a linear part and two logistic gates:
with .
Linear Autoregression (LAR). This process is an AR(6) fit to the annual sunspot series [9]:
with . The process is linear, so only scales the series. It shows the cost of a flexible decoder when the true process is linear.
Training. Each training sample has short series. For each process, we discard the first 300 simulated values. We train each strategy for 2,500 to 3,000 steps on each sample. We use independent samples for the bias and variance estimates, and we report the horizons up to . The series length changes as in Ben Taieb and Atiya [5].
NLAR MSE Decomposition

Figure C.8. MSE decomposition on the NLAR process [5] across training-series lengths and horizons . Direct and multi-step strategies reproduce the classical bias–variance trade-off, while Continuous Horizon (CH) attains the lowest total error, matching multi-step at short lengths and pulling ahead as increases.
STAR MSE Decomposition

Figure C.9. MSE decomposition on the STAR process [5], in the same layout as Figure C.8. The recurrent strategy exhibits visible bias from rollout limitations on the smooth-transition nonlinearity, whereas all other strategies—including Continuous Horizon (CH)—are near-optimal on this nearly deterministic process.
Unseen Horizons Experiment Details
Data. The experiment in Section 5 uses zero-mean Gaussian processes. A Gaussian process has a value at each real time, so it gives true targets at fractional and unseen horizons. We use three kernels, with :
Table D.5 gives the kernel and the parameters of each regime. We add observation noise with to each series.
| Regime | Kernel | |||
| Mean-reverting | 3 | 12 | 12 | |
| Aperiodic | – | – | 16 | |
| Periodic | 3 | 12 | – | |
| QP (slow) | 3 | 12 | 48 | |
| QP (fast) | 3 | 12 | 24 |
Table D.5. Kernels and parameters of the five Gaussian-process regimes.
Exact Conditional Mean. For a context , the posterior Gaussian-process gives the exact conditional mean at each real horizon . The posterior includes the same observation noise, so the regret of Eq. (14) is exact. The forecasters do not know the kernel.
Models. All strategies use the same two-layer encoder in a context window of 16 observations. The CH decoder is the branch–trunk expansion of Eq. (16). MS uses one shared output layer for all horizons, DIR uses one output layer for each horizon, and REC applies a one-step output layer recursively. MS and CH have approximately 5,000 parameters each.
Training and evaluation. Each training sample has 64 series. We train each strategy for 8,000 steps on the integer horizons . We repeat the training on independent samples. All strategies use pinball loss on quantile levels , and the point forecast is the median. For a Gaussian posterior, the median is equal to the conditional mean . We evaluate the integer horizons seen, the half-integer horizons, and the integer horizons .
Qualitative Forecast Exploration
Figure E.10. Forecasts of an oscillating Gaussian Process signal. Black: GP truth. Blue: CH. Red dashed: MS with linear interpolation, with markers at its output horizons. Left to right: coarse, intermediate, and full MS grid.
Here we complement the accuracy results of § 5, with a qualitative exploration of the outputs of discrete and CH forecast strategy.
The strategies MS, REC, DIR produce forecasts discretely with the output grid . For any other horizon , to obtain the forecast, we must interpolate between two adjacent horizons.
Horizons therefore in the interpolation regime can be thought of as interpolant knots. The forecast accuracy between the knots depends on the distance between them. For an output grid with intervals on , .
For linear interpolation, the error between two knots is not more than times the maximum curvature of the true forecast. Thus, if the grid has half the number of horizons, doubles, and the error bound becomes four times larger [8]. A discrete forecast strategy can show an oscillation only if the distance between knots is less than half of the target signal period [56]. If the horizon distance is larger, the discrete strategies cannot contain the cycles. Then, no interpolation method can find the cycle again, and the result is aliasing. The coarse panel of Figure E.10 shows this result: the MS interpolant is an almost straight line.
The strategy CH bypasses discrete horizons. It learns the forecast as a continuous function of , so the interpolation error of the MS grid does not occur. But CH also trains only on integer horizons. Its accuracy between these horizons depends on the smoothness of the learned horizon basis functions , not on the density of the grid.
Horizon-Basis Ablation
The horizon basis lets CH forecast at unseen horizons. All variants use the same encoder, parameter budget, training schedule, and branch coefficients. Only changes.
Protocol. We use the Gaussian-process data of Appendix D, with 48 series in each training sample, 3,000 training steps, and replicates. Some kernel parameters are different from Table D.5. The mean-reverting regime uses with , and the periodic regime uses . The slow and the fast quasi-periodic regimes use and . The metric is the RMSE of Eq. (13).
Basis blocks. The basis uses five blocks. Each block is a function of the horizon , or a constant:
The blocks and give a polynomial trend. The Fourier block has fixed or learnable frequencies . The table block interpolates a learned embedding table , with . For , it keeps the last value . The block is a learned frequency embedding, and it does not depend on .
Composite basis. The basis is a linear combination of the blocks:
Each matrix is learned, with the dimension of block . A variant removes block with . The basis does not depend on the history, so it is separable, as in Eq. (16).
Non-separable variants. Two variants use the encoder state in the basis. The gated variant multiplies the quadratic and the Fourier blocks by gates from the history, with :
The per-series variant uses the frequency embedding () and replaces with , which the encoder gives. In both variants, the basis depends on the history, so it is not separable.
Ablations. Each ablation changes one part of Eq. (F.1) or Eq. (F.2). The anchor uses all blocks except the frequency embedding (), with learnable and . The ablations are:
Composition (Table F.6): all seven combinations of the polynomial (lin and quad), the Fourier, and the table blocks, with and for each unused block.
Basis Dimensionality (Table F.7): .
Frequency representation (Table F.8): fixed or learnable , the frequency embedding with and without the Fourier block (), the per-series embedding, and the gate.
In Table F.8, the Sep. column shows the separable variants, to which Appendix § 9 and Appendix B apply. MS and DIR are not in these tables, because they cannot forecast at unseen horizons.
Results. On the seen and the interpolation horizons, all variants have almost the same error. The exceptions are the polynomial-only basis and on the periodic regime. On the extrapolation horizons, the variants are different:
No composition is the best in all regimes. The table-only basis is the best in three regimes.
The polynomial-only basis has the highest error in three regimes.
From to , the error marginally decreases by 5% or less.
Without the Fourier block, the frequency embedding has a higher error on the periodic regimes.
The non-separable variants have the lowest error in four regimes. Their advantage over the best separable variant is only 1.3% or less.
| Regime | Basis composition | Seen | Interp. | Extrap. |
| Mean-reverting | polynomial only | 1.092 | 1.100 | 1.610 |
| Mean-reverting | Fourier only | 1.092 | 1.101 | 1.226 |
| Mean-reverting | table only | 1.091 | 1.099 | 1.153 |
| Mean-reverting | polynomial + Fourier | 1.094 | 1.103 | 1.263 |
| Mean-reverting | polynomial + table | 1.091 | 1.099 | 1.333 |
| Mean-reverting | Fourier + table | 1.091 | 1.099 | 1.188 |
| Mean-reverting | full composite (anchor) | 1.092 | 1.101 | 1.217 |
| Aperiodic | polynomial only | 0.715 | 0.713 | 1.077 |
| Aperiodic | Fourier only | 0.715 | 0.713 | 1.085 |
| Aperiodic | table only | 0.715 | 0.712 | 1.030 |
| Aperiodic | polynomial + Fourier | 0.715 | 0.713 | 1.057 |
| Aperiodic | polynomial + table | 0.715 | 0.712 | 1.054 |
| Aperiodic | Fourier + table | 0.715 | 0.713 | 1.065 |
| Aperiodic | full composite (anchor) | 0.715 | 0.712 | 1.054 |
| Periodic | polynomial only | 0.674 | 0.658 | 5.339 |
| Periodic | Fourier only | 0.643 | 0.638 | 0.740 |
| Periodic | table only | 0.642 | 0.633 | 0.925 |
| Periodic | polynomial + Fourier | 0.642 | 0.637 | 0.762 |
| Periodic | polynomial + table | 0.642 | 0.633 | 1.138 |
| Periodic | Fourier + table | 0.642 | 0.636 | 0.691 |
| Periodic | full composite (anchor) | 0.641 | 0.635 | 0.729 |
| Quasi-periodic (slow) | polynomial only | 0.700 | 0.698 | 1.391 |
| Quasi-periodic (slow) | Fourier only | 0.683 | 0.683 | 0.943 |
| Quasi-periodic (slow) | table only | 0.690 | 0.688 | 0.935 |
| Quasi-periodic (slow) | polynomial + Fourier | 0.683 | 0.683 | 0.929 |
| Quasi-periodic (slow) | polynomial + table | 0.691 | 0.689 | 0.989 |
| Quasi-periodic (slow) | Fourier + table | 0.683 | 0.683 | 0.926 |
| Quasi-periodic (slow) | full composite (anchor) | 0.683 | 0.682 | 0.919 |
| Quasi-periodic (fast) | polynomial only | 0.983 | 0.984 | 1.228 |
| Quasi-periodic (fast) | Fourier only | 0.981 | 0.983 | 1.280 |
| Quasi-periodic (fast) | table only | 0.981 | 0.982 | 1.175 |
| Quasi-periodic (fast) | polynomial + Fourier | 0.982 | 0.984 | 1.210 |
| Quasi-periodic (fast) | polynomial + table | 0.981 | 0.983 | 1.218 |
| Quasi-periodic (fast) | Fourier + table | 0.980 | 0.982 | 1.242 |
| Quasi-periodic (fast) | full composite (anchor) | 0.982 | 0.983 | 1.202 |
Table F.6. Ablation of the basis composition. RMSE on the seen, interpolation, and extrapolation horizons of the five regimes. The anchor is the full composite basis with learnable frequencies and . Bold shows the lowest extrapolation error.
| Regime | Seen | Interp. | Extrap. | |
| Mean-reverting | 1.091 | 1.102 | 1.267 | |
| 1.093 | 1.103 | 1.234 | ||
| 1.093 | 1.102 | 1.229 | ||
| 1.092 | 1.101 | 1.217 | ||
| 1.092 | 1.100 | 1.203 | ||
| Aperiodic | 0.716 | 0.716 | 1.059 | |
| 0.715 | 0.713 | 1.054 | ||
| 0.715 | 0.713 | 1.055 | ||
| 0.715 | 0.712 | 1.054 | ||
| 0.715 | 0.712 | 1.052 | ||
| Periodic | 0.665 | 0.704 | 1.529 | |
| 0.641 | 0.640 | 0.748 | ||
| 0.641 | 0.638 | 0.740 | ||
| 0.641 | 0.635 | 0.729 | ||
| 0.643 | 0.636 | 0.713 | ||
| Quasi-periodic (slow) | 0.699 | 0.701 | 0.971 | |
| 0.684 | 0.684 | 0.934 | ||
| 0.683 | 0.683 | 0.930 | ||
| 0.683 | 0.682 | 0.919 | ||
| 0.683 | 0.682 | 0.915 | ||
| Quasi-periodic (fast) | 0.982 | 0.987 | 1.241 | |
| 0.981 | 0.983 | 1.213 | ||
| 0.981 | 0.983 | 1.201 | ||
| 0.982 | 0.983 | 1.202 | ||
| 0.981 | 0.982 | 1.199 |
Table F.7. Ablation of the number of basis functions, with the full composite basis. RMSE on the seen, interpolation, and extrapolation horizons of the five regimes.
| Regime | Frequency representation | Seen | Interp. | Extrap. | Sep. |
| Mean-reverting | fixed Fourier dictionary | 1.092 | 1.101 | 1.234 | ✓ |
| Mean-reverting | learnable Fourier (anchor) | 1.092 | 1.101 | 1.217 | ✓ |
| Mean-reverting | frequency embedding, shared | 1.093 | 1.102 | 1.213 | ✓ |
| Mean-reverting | frequency embedding, per-series | 1.092 | 1.100 | 1.204 | ✗ |
| Mean-reverting | frequency embedding only (no Fourier) | 1.091 | 1.099 | 1.312 | ✓ |
| Mean-reverting | gated composite | 1.092 | 1.100 | 1.212 | ✗ |
| Aperiodic | fixed Fourier dictionary | 0.715 | 0.713 | 1.047 | ✓ |
| Aperiodic | learnable Fourier (anchor) | 0.715 | 0.712 | 1.054 | ✓ |
| Aperiodic | frequency embedding, shared | 0.715 | 0.713 | 1.049 | ✓ |
| Aperiodic | frequency embedding, per-series | 0.716 | 0.713 | 1.052 | ✗ |
| Aperiodic | frequency embedding only (no Fourier) | 0.715 | 0.712 | 1.046 | ✓ |
| Aperiodic | gated composite | 0.715 | 0.713 | 1.036 | ✗ |
| Periodic | fixed Fourier dictionary | 0.642 | 0.637 | 0.746 | ✓ |
| Periodic | learnable Fourier (anchor) | 0.641 | 0.635 | 0.729 | ✓ |
| Periodic | frequency embedding, shared | 0.642 | 0.635 | 0.703 | ✓ |
| Periodic | frequency embedding, per-series | 0.642 | 0.635 | 0.701 | ✗ |
| Periodic | frequency embedding only (no Fourier) | 0.642 | 0.633 | 1.098 | ✓ |
| Periodic | gated composite | 0.642 | 0.635 | 0.733 | ✗ |
| Quasi-periodic (slow) | fixed Fourier dictionary | 0.683 | 0.683 | 0.917 | ✓ |
| Quasi-periodic (slow) | learnable Fourier (anchor) | 0.683 | 0.682 | 0.919 | ✓ |
| Quasi-periodic (slow) | frequency embedding, shared | 0.683 | 0.682 | 0.914 | ✓ |
| Quasi-periodic (slow) | frequency embedding, per-series | 0.684 | 0.683 | 0.918 | ✗ |
| Quasi-periodic (slow) | frequency embedding only (no Fourier) | 0.690 | 0.688 | 0.981 | ✓ |
| Quasi-periodic (slow) | gated composite | 0.685 | 0.684 | 0.938 | ✗ |
| Quasi-periodic (fast) | fixed Fourier dictionary | 0.981 | 0.983 | 1.195 | ✓ |
| Quasi-periodic (fast) | learnable Fourier (anchor) | 0.982 | 0.983 | 1.202 | ✓ |
| Quasi-periodic (fast) | frequency embedding, shared | 0.981 | 0.983 | 1.199 | ✓ |
| Quasi-periodic (fast) | frequency embedding, per-series | 0.980 | 0.982 | 1.200 | ✗ |
| Quasi-periodic (fast) | frequency embedding only (no Fourier) | 0.981 | 0.983 | 1.205 | ✓ |
| Quasi-periodic (fast) | gated composite | 0.982 | 0.983 | 1.180 | ✗ |
Table F.8. Ablation of the frequency representation. RMSE on the seen, interpolation, and extrapolation horizons of the five regimes. A check mark in the Sep. column shows a separable basis. Bold shows the lowest extrapolation error in each regime.
Dataset Details
| Frequency | Seasonality | Horizon | Series | Min Length | Max Length | % Erratic | |
| M1 | Monthly | 12 | 18 | 617 | 48 | 150 | 0 |
| M1 | Quarterly | 4 | 8 | 203 | 18 | 114 | 0 |
| M1 | Yearly | 1 | 6 | 181 | 15 | 58 | 0 |
| M3 | Other | 4 | 8 | 174 | 71 | 104 | 0 |
| M3 | Monthly | 12 | 18 | 1428 | 66 | 144 | 2 |
| M3 | Quarterly | 4 | 8 | 756 | 24 | 72 | 1 |
| M3 | Yearly | 1 | 6 | 645 | 20 | 47 | 10 |
| M4 | Hourly | 24 | 48 | 414 | 748 | 1008 | 17 |
| M4 | Daily | 1 | 14 | 4,227 | 107 | 9933 | 2 |
| M4 | Weekly | 1 | 13 | 359 | 93 | 2610 | 16 |
| M4 | Monthly | 12 | 18 | 48,000 | 60 | 2812 | 6 |
| M4 | Quarterly | 4 | 8 | 24,000 | 24 | 874 | 11 |
| M4 | Yearly | 1 | 6 | 23,000 | 19 | 841 | 18 |
| Tourism | Monthly | 12 | 18 | 366 | 91 | 333 | 51 |
| Tourism | Quarterly | 4 | 8 | 427 | 30 | 130 | 39 |
| Tourism | Yearly | 1 | 6 | 518 | 11 | 47 | 23 |
Table G.9. Summary of forecasting datasets used in our main empirical study.
M1. The M1 competition [40] has 1,001 monthly, quarterly, and yearly series from demography, industry, and economics. In this competition, simple methods such as ETS [25] were frequently more accurate than complex methods. The M1 competition initiated a research legacy that emphasized forecast accuracy.
M3. The M3 competition [41] has 3,003 series of business, demography, finance, and economics. All values are positive, and no series is intermittent [58]. The Theta method [30] had the best overall accuracy.
M4. The M4 competition [42] has 100,000 series at six frequencies, from demography, finance, industry, and economics. It added the evaluation of prediction intervals. ESRNN [57], a hybrid of exponential smoothing and a neural network, won the competition. M4 also made cross-learning and global models popular [55].
Tourism. The Tourism dataset [3] has 1,311 monthly, quarterly, and yearly series of tourism demand. This competition was the first to use the mean absolute scaled error (MASE). Of all series, 36% are erratic or intermittent. At the yearly frequency, few methods were more accurate than the Naïve1 method.
Training Methodology and Hyperparameters
Network Hyperparameters
We train MQForecaster varying encoders between LSTM, CNN, Transformer, and S4 modules. We train models with hyperparameter tuning using Optuna. Using 5 trials, we train individual models and select the hyperparameters that minimize the validation RMSE. Training times mostly depend on the architecture; however, we restrict the SGD training steps to 45K per architecture. Tables H.10 and H.11–H.14 list the shared hyperparameters and encoder-specific hyperparameters, respectively.
| Hyperparameter | Values |
| Single GPU SGD Batch Size{sup}* | |
| Initial learning rate | loguniform(1e-4, 5e-2) |
| Maximum Training steps | 45,000 |
| Learning rate decay | 0.1 |
| Learning rate step size | 15,000 |
| Scaler type | Standard |
| H-Agnostic Decoder Dimension | 100 |
| H-Specific Decoder Dimension | 20 |
Table H.10. Shared hyperparameters across encoder types
| Hyperparameter | Values |
| Main Activation Functions | Sigmoid, tanh |
| Encoder Dimension | |
| Dilations |
Table H.11. LSTM
| Hyperparameter | Values |
| Main Activation Function | ReLU |
| Encoder Dimension | |
| Temporal Convolution Kernel Size | |
| Temporal Convolution Dilations |
Table H.12. CNN
| Hyperparameter | Values |
| Main Activation Functions | Sigmoid, tanh |
| Encoder Dimension | |
| Projection Dimension | |
| Input Patch Lengths | |
| Attention Dilations | |
| Attention Dropout | 0.1 |
| Number of Attention Heads | 4 |
Table H.13. Transformer
| Hyperparameter | Values |
| Main Activation Function | GELU |
| Encoder Dimension | |
| Projection Hidden Size |
Table H.14. State Space S4
Parameter-Matched Strategy Comparison
We compare the strategies at the same number of parameters, so that an accuracy difference comes from the horizon strategy, not from the model size. The decoders of MS and CH scale differently with the horizon . The MS decoder has one output block for each horizon, so its size increases with . The CH decoder uses a basis with a fixed size. Without matching, MS has more parameters than CH, and the difference increases with .
To match parameters, we marginally decrease the width of the encoder output for MS, so that the totals agree. For example, the width is 32 for CH and 19 for MS with the CNN encoder. Then the totals are different by 3.1% or less (Table H.15). We count the parameters directly, because the same width gives different counts for different decoders.
| Model | LSTM | CNN | Transformer | S4 | ||||||||||||
| CH | MS | DIR* | REC | CH | MS | DIR* | REC | CH | MS | DIR* | REC | CH | MS | DIR* | REC | |
| Global model | 78,482 | 77,865 | 7,178 | 78,482 | 63,154 | 65,125 | 9,477 | 63,154 | 81,746 | 80,868 | 8,641 | 81,746 | 85,714 | 86,674 | 13,961 | 85,714 |
Table H.15. Trainable MQForecaster parameters for each encoder and strategy. A dash marks a strategy not trained with that encoder; counts are functional (CH basis/embeddings disabled for DIR and MS). CH, MS, and REC are single global models. ∗DIR trains one model per horizon; the listed count is per horizon, split as encoder + decoder (CNN: 8,006+1,174). The encoder matches the global model; the decoder is of the global decoder, so over horizons the DIR decoders sum to one global decoder (, vs. 19,221) while the encoder is replicated ().
Regret-Based Dataset Sampling for Cross-Learning
We train one global model on the 16 datasets of § 6, by cross-learning [55]. The datasets are very different in size, so proportional sampling gives the small datasets few gradient updates. Thus, we change the sampling probability of each dataset during training. A dataset gets a larger probability when its validation error is high relative to a reference for that dataset. This method is a variant of group distributionally robust optimization [53] and of data-mixture reweighting of Xie et al. [65].
The Continuous-Horizon Learning Objective
MS Learning Objective. Theorem 2 shows the benefit of training on fractional horizons. But the M-series experiments compare the inductive bias of the decoders, with the learning objective fixed. Thus, we train CH on the integer horizons , with the traditional MS objective. The uniform measure of Eq. (6) then becomes the uniform measure on the integer grid:
The loss is the MSE. MS minimizes Eq. (H.1) with all horizons in each training step. For CH, we used the same loss and the same pooled data, but we sampled the horizons.
CH Learning Objective. In each training step, we sample a random subset of the horizons for the full mini-batch. We drop each horizon independently with probability 0.2 and we keep at least 10 horizons. Each horizon has the same probability to be in , so
Thus, a training step with minimizes Eq. (H.1) in expectation. In our experiments, the number of sampled horizons changed the validation accuracy by less than 2%.
Additional Point Forecast Metrics
We complement the relAvgRMSE of Table 3 with the two absolute-error metrics most used in the M-competitions, so that our results can be read against previous work on these datasets. Both metrics use the mean forecast and the pool over all series , the forecast creation dates and the horizons of the temporal cross-validation of § 6.
| LSTM CH | LSTM MS | LSTM DIR | LSTM REC | CNN CH | CNN MS | CNN DIR | CNN REC | Transformer CH | Transformer MS | Transformer DIR | Transformer REC | S4 CH | S4 MS | S4 DIR | S4 REC | StatsForecast ETS | StatsForecast ARIMA | ||
| M1 | M | 0.8838 (0.1518) | 0.6001 (0.0921) | 0.2027 (0.0007) | 0.8644 (0.1172) | 0.1914 (0.0009) | 0.1945 (0.0007) | 0.1974 (0.0027) | 0.2308 (0.0064) | 0.1953 (0.0041) | 0.2104 (0.0050) | 0.2126 (0.0023) | 0.2335 (0.0073) | 0.1954 (0.0004) | 0.1975 (0.0013) | 0.2077 (0.0029) | 0.2315 (0.0047) | 0.1693 (-) | 0.1807 (-) |
| M1 | Q | 1.0103 (0.1186) | 0.5547 (0.0914) | 0.1508 (0.0023) | 0.9659 (0.1832) | 0.1410 (0.0006) | 0.1369 (0.0010) | 0.1431 (0.0018) | 0.1479 (0.0017) | 0.1456 (0.0028) | 0.1505 (0.0062) | 0.1615 (0.0046) | 0.1574 (0.0041) | 0.1456 (0.0010) | 0.1425 (0.0021) | 0.1533 (0.0011) | 0.1596 (0.0053) | 0.1624 (-) | 0.1881 (-) |
| M1 | Y | 1.2673 (0.2036) | 1.0504 (0.2436) | 0.1622 (0.0021) | 1.2674 (0.2578) | 0.1601 (0.0022) | 0.1490 (0.0026) | 0.1504 (0.0023) | 0.1565 (0.0041) | 0.1690 (0.0019) | 0.1773 (0.0051) | 0.1761 (0.0072) | 0.1616 (0.0045) | 0.1626 (0.0025) | 0.1612 (0.0023) | 0.1785 (0.0042) | 0.1674 (0.0041) | 0.2188 (-) | 0.1908 (-) |
| M3 | O | 0.4495 (0.1494) | 0.2826 (0.0485) | 0.0469 (0.0013) | 0.3481 (0.0925) | 0.0360 (0.0003) | 0.0416 (0.0003) | 0.0471 (0.0013) | 0.0505 (0.0023) | 0.0477 (0.0015) | 0.0527 (0.0031) | 0.0846 (0.0267) | 0.0572 (0.0064) | 0.0417 (0.0008) | 0.0408 (0.0030) | 0.0510 (0.0043) | 0.0593 (0.0036) | 0.0433 (-) | 0.0438 (-) |
| M3 | M | 0.5901 (0.1733) | 0.4287 (0.0836) | 0.1583 (0.0008) | 0.5309 (0.1017) | 0.1486 (0.0002) | 0.1510 (0.0009) | 0.1510 (0.0007) | 0.1887 (0.0029) | 0.1542 (0.0026) | 0.1605 (0.0017) | 0.1663 (0.0038) | 0.1904 (0.0068) | 0.1488 (0.0003) | 0.1487 (0.0011) | 0.1513 (0.0014) | 0.1854 (0.0036) | 0.1627 (-) | 0.1645 (-) |
| M3 | Q | 0.8036 (0.1348) | 0.3693 (0.0364) | 0.1073 (0.0014) | 0.7295 (0.1438) | 0.0986 (0.0010) | 0.0985 (0.0006) | 0.1039 (0.0010) | 0.1099 (0.0017) | 0.1047 (0.0028) | 0.1122 (0.0062) | 0.1214 (0.0046) | 0.1147 (0.0041) | 0.0981 (0.0010) | 0.0920 (0.0021) | 0.1115 (0.0011) | 0.1156 (0.0053) | 0.1014 (-) | 0.1020 (-) |
| M3 | Y | 1.1225 (0.1953) | 0.8359 (0.1592) | 0.1170 (0.0020) | 1.1435 (0.1755) | 0.1658 (0.0018) | 0.1573 (0.0019) | 0.1508 (0.0018) | 0.1677 (0.0043) | 0.1624 (0.0046) | 0.1773 (0.0058) | 0.1753 (0.0073) | 0.1687 (0.0049) | 0.1658 (0.0024) | 0.1547 (0.0028) | 0.1680 (0.0042) | 0.1769 (0.0048) | 0.1971 (-) | 0.1953 (-) |
| M4 | H | 0.8672 (0.1765) | 0.6679 (0.1299) | 0.2222 (0.0022) | 0.8421 (0.1158) | 0.2236 (0.0019) | 0.2228 (0.0028) | 0.2327 (0.0034) | 0.3135 (0.0094) | 0.2084 (0.0018) | 0.2328 (0.0174) | 0.2972 (0.0039) | 0.3703 (0.0170) | 0.2274 (0.0008) | 0.1921 (0.0014) | 0.1788 (0.0031) | 0.3492 (0.0066) | 0.2336 (-) | 0.1426 (-) |
| M4 | D | 0.2972 (0.1323) | 0.1540 (0.0170) | 0.0319 (0.0039) | 0.1878 (0.1878) | 0.0302 (0.0002) | 0.0303 (0.0003) | 0.0304 (0.0004) | 0.0934 (0.0094) | 0.0322 (0.0019) | 0.0334 (0.0174) | 0.0334 (0.0034) | 0.0370 (0.0170) | 0.0308 (0.0008) | 0.0301 (0.0014) | 0.0318 (0.0031) | 0.0371 (0.0066) | 0.0280 (-) | 0.0285 (-) |
| M4 | W | 0.4587 (0.1493) | 0.2629 (0.0558) | 0.0979 (0.0014) | 0.3781 (0.0862) | 0.0887 (0.0007) | 0.0930 (0.0010) | 0.0929 (0.0019) | 0.0942 (0.0017) | 0.0956 (0.0017) | 0.0974 (0.0126) | 0.1003 (0.0025) | 0.1052 (0.0054) | 0.0925 (0.0013) | 0.0914 (0.0014) | 0.0971 (0.0031) | 0.1031 (0.0066) | 0.0903 (-) | 0.0878 (-) |
| M4 | M | 0.5515 (0.1717) | 0.5817 (0.1170) | 0.1445 (0.0044) | 0.4866 (0.0866) | 0.1397 (0.0013) | 0.1402 (0.0012) | 0.1397 (0.0017) | 0.1597 (0.0017) | 0.1370 (0.0017) | 0.1428 (0.0128) | 0.1533 (0.0025) | 0.1785 (0.0170) | 0.1357 (0.0013) | 0.1401 (0.0014) | 0.1394 (0.0031) | 0.1726 (0.0066) | 0.1321 (-) | 0.1381 (-) |
| M4 | Q | 0.6464 (0.1166) | 0.2833 (0.0199) | 0.1199 (0.0017) | 0.5502 (0.1442) | 0.1124 (0.0011) | 0.1117 (0.0017) | 0.1117 (0.0017) | 0.1220 (0.0017) | 0.1238 (0.0025) | 0.1291 (0.0127) | 0.1291 (0.0025) | 0.1327 (0.0170) | 0.1152 (0.0012) | 0.1174 (0.0014) | 0.1176 (0.0031) | 0.1273 (0.0066) | 0.1045 (-) | 0.1070 (-) |
| M4 | Y | 1.1435 (0.1617) | 0.7597 (0.0867) | 0.1561 (0.0021) | 1.1286 (0.1599) | 0.1503 (0.0014) | 0.1426 (0.0019) | 0.1424 (0.0019) | 0.1524 (0.0022) | 0.1466 (0.0017) | 0.1528 (0.0170) | 0.1602 (0.0025) | 0.1473 (0.0170) | 0.1555 (0.0013) | 0.1421 (0.0014) | 0.1545 (0.0031) | 0.1585 (0.0066) | 0.1546 (-) | 0.1616 (-) |
| Tourism | M | 1.0228 (0.1839) | 0.7346 (0.1126) | 0.3498 (0.0098) | 1.0463 (0.1921) | 0.3180 (0.0044) | 0.3200 (0.0033) | 0.3198 (0.0056) | 0.4517 (0.0214) | 0.3605 (0.0063) | 0.3727 (0.0055) | 0.3766 (0.0071) | 0.4672 (0.0291) | 0.3159 (0.0022) | 0.3024 (0.0017) | 0.3089 (0.0049) | 0.4662 (0.0133) | 0.2220 (-) | 0.2206 (-) |
| Tourism | Q | 0.9581 (0.1454) | 0.4654 (0.0601) | 0.2833 (0.0023) | 0.9156 (0.1916) | 0.2221 (0.0037) | 0.2324 (0.0024) | 0.2588 (0.0058) | 0.2839 (0.0190) | 0.2947 (0.0049) | 0.2990 (0.0029) | 0.3021 (0.0071) | 0.2975 (0.0221) | 0.2179 (0.0023) | 0.2166 (0.0022) | 0.2585 (0.0055) | 0.2856 (0.0133) | 0.1797 (-) | 0.1785 (-) |
| Tourism | Y | 1.3570 (0.2633) | 1.0668 (0.1969) | 0.2241 (0.0033) | 1.1680 (0.2712) | 0.2076 (0.0026) | 0.2179 (0.0037) | 0.2016 (0.0033) | 0.2086 (0.0033) | 0.1865 (0.0029) | 0.1975 (0.0028) | 0.2052 (0.0049) | 0.2059 (0.0133) | 0.2077 (0.0023) | 0.1973 (0.0022) | 0.2052 (0.0055) | 0.2041 (0.0133) | 0.2870 (-) | 0.3030 (-) |
Table I.16. Empirical evaluation of point forecasts. Symmetric Mean Absolute Percentage Error (sMAPE, Eq. (I.(1)), averaged over 5 runs. Lower measurements are preferred. The methods without standard deviation have deterministic solutions. For each encoder we compare four forecasting strategies: our Continuous Horizon (CH), multi-step (MS), direct (DIR), and recurrent (REC).
The symmetric mean absolute percentage error is
| LSTM | CNN | Transformer | S4 | StatsForecast | |||||||||||||||
| Dataset | Freq | CH | MS | DIR | REC | CH | MS | DIR | REC | CH | MS | DIR | REC | CH | MS | DIR | REC | ETS | ARIMA |
| M1 | M | 0.8129 (0.0062) | 0.8817 (0.0054) | 0.9279 (0.0045) | 0.9403 (0.0168) | 0.8295 (0.0048) | 0.9047 (0.0080) | 0.9196 (0.0217) | 1.0171 (0.0439) | 0.8971 (0.0231) | 0.9421 (0.0097) | 0.9732 (0.0225) | 1.0009 (0.0101) | 0.9019 (0.0032) | 0.8957 (0.0097) | 0.9437 (0.0282) | 1.0026 (0.0143) | 0.7319 (-) | 0.7660 (-) |
| Q | 0.9274 (0.0268) | 0.9426 (0.0140) | 0.9765 (0.0262) | 0.9207 (0.0211) | 0.9095 (0.0047) | 0.9353 (0.0136) | 0.9897 (0.0204) | 1.2168 (0.1247) | 1.0114 (0.0771) | 1.0400 (0.0244) | 1.0644 (0.0465) | 0.9591 (0.0228) | 0.9560 (0.0201) | 0.9363 (0.0144) | 1.0444 (0.0143) | 0.9269 (0.0204) | 0.8980 (-) | 0.8899 (-) | |
| Y | 0.8187 (0.0209) | 0.7250 (0.0253) | 0.7733 (0.0274) | 0.9049 (0.0833) | 0.7830 (0.0132) | 0.8005 (0.0391) | 0.6960 (0.0475) | 0.7472 (0.0430) | 0.6603 (0.0319) | 0.8033 (0.0853) | 0.8408 (0.0841) | 0.7948 (0.0561) | 0.7373 (0.0180) | 0.8117 (0.0378) | 0.7340 (0.0509) | 0.8250 (0.0534) | 0.7838 (-) | 0.7534 (-) | |
| M3 | O | 0.2018 (0.0021) | 0.2106 (0.0219) | 0.2080 (0.0065) | 0.2241 (0.0137) | 0.1552 (0.0014) | 0.1859 (0.0033) | 0.2074 (0.0068) | 0.3381 (0.0660) | 0.2191 (0.0118) | 0.2490 (0.0097) | 0.3886 (0.1386) | 0.2517 (0.0209) | 0.1786 (0.0018) | 0.1754 (0.0118) | 0.2250 (0.0163) | 0.2634 (0.0159) | 0.6915 (-) | 0.7040 (-) |
| M | 0.6646 (0.0049) | 0.6846 (0.0077) | 0.7538 (0.0068) | 0.8394 (0.0257) | 0.6967 (0.0012) | 0.7135 (0.0032) | 0.7131 (0.0034) | 1.0855 (0.0458) | 0.7364 (0.0182) | 0.7654 (0.0109) | 0.7978 (0.0207) | 0.9903 (0.0668) | 0.7033 (0.0022) | 0.7012 (0.0053) | 0.7147 (0.0040) | 0.9417 (0.0254) | 0.8555 (-) | 0.8622 (-) | |
| Q | 0.7772 (0.0097) | 0.7874 (0.0118) | 0.8512 (0.0120) | 0.8999 (0.0217) | 0.7719 (0.0090) | 0.7832 (0.0091) | 0.8322 (0.0147) | 1.0647 (0.0942) | 0.8695 (0.0523) | 0.9260 (0.0376) | 0.9670 (0.0166) | 0.9168 (0.0434) | 0.7687 (0.0090) | 0.7247 (0.0081) | 0.8996 (0.0044) | 0.9328 (0.0118) | 0.9277 (-) | 0.9257 (-) | |
| Y | 0.8905 (0.0141) | 0.7684 (0.0093) | 0.8637 (0.0104) | 0.9120 (0.0521) | 0.8189 (0.0118) | 0.7786 (0.0093) | 0.7543 (0.0143) | 0.9559 (0.0705) | 0.8759 (0.0725) | 0.8620 (0.0455) | 0.8649 (0.0300) | 0.8243 (0.0249) | 0.8637 (0.0119) | 0.7666 (0.0105) | 0.8610 (0.0106) | 0.8814 (0.0082) | 0.9716 (-) | 0.9868 (-) | |
| M4 | H | 0.1617 (0.0013) | 0.1703 (0.0052) | 0.3738 (0.0031) | 0.6322 (0.0592) | 0.2715 (0.0015) | 0.4464 (0.0047) | 0.4665 (0.0063) | 0.6923 (0.0592) | 0.2363 (0.0156) | 0.4631 (0.0370) | 0.6826 (0.0295) | 0.9682 (0.1702) | 0.3018 (0.0138) | 0.2829 (0.0062) | 0.2617 (0.0058) | 0.5793 (0.0393) | 0.4148 (-) | 0.1899 (-) |
| D | 0.7924 (0.0031) | 0.7412 (0.0130) | 0.8025 (0.0059) | 0.8704 (0.0393) | 0.7872 (0.0075) | 0.7749 (0.0077) | 0.7940 (0.0099) | 0.8499 (0.0557) | 0.8241 (0.0204) | 0.7998 (0.0171) | 0.8317 (0.0295) | 0.9598 (0.0594) | 0.7759 (0.0276) | 0.7461 (0.0091) | 0.8034 (0.0178) | 0.9363 (0.0377) | 0.9787 (-) | 0.9892 (-) | |
| W | 0.7441 (0.0131) | 0.6873 (0.0077) | 0.6713 (0.0113) | 0.7483 (0.0748) | 0.7249 (0.0049) | 0.7357 (0.0077) | 0.7323 (0.0074) | 0.8055 (0.0582) | 0.7847 (0.0178) | 0.7815 (0.0208) | 0.7907 (0.0297) | 0.8394 (0.0585) | 0.7427 (0.0074) | 0.7240 (0.0062) | 0.7747 (0.0178) | 0.8377 (0.0359) | 0.9470 (-) | 0.9179 (-) | |
| M | 0.6707 (0.0069) | 0.6889 (0.0079) | 0.7013 (0.0078) | 0.7915 (0.0615) | 0.6857 (0.0047) | 0.6847 (0.0077) | 0.6791 (0.0075) | 0.9764 (0.0836) | 0.6821 (0.0191) | 0.7108 (0.0178) | 0.7503 (0.0306) | 0.9730 (0.0972) | 0.6589 (0.0074) | 0.6818 (0.0067) | 0.6799 (0.0178) | 0.9073 (0.0395) | 0.8897 (-) | 0.9006 (-) | |
| Q | 0.7112 (0.0035) | 0.7054 (0.0084) | 0.7606 (0.0115) | 0.7661 (0.0248) | 0.7034 (0.0052) | 0.7037 (0.0064) | 0.7162 (0.0155) | 0.8519 (0.0558) | 0.8109 (0.0197) | 0.7952 (0.0166) | 0.8268 (0.0306) | 0.8831 (0.0704) | 0.7290 (0.0066) | 0.7425 (0.0038) | 0.7624 (0.0150) | 0.8340 (0.0283) | 0.8727 (-) | 0.9168 (-) | |
| Y | 0.7954 (0.0031) | 0.6783 (0.0061) | 0.7837 (0.0051) | 0.8242 (0.0652) | 0.7364 (0.0077) | 0.7151 (0.0070) | 0.7143 (0.0155) | 0.8398 (0.0558) | 0.7959 (0.0597) | 0.7966 (0.0575) | 0.7956 (0.0294) | 0.7277 (0.0254) | 0.7744 (0.0066) | 0.7088 (0.0038) | 0.7888 (0.0049) | 0.8042 (0.0061) | 0.8700 (-) | 0.9138 (-) | |
| Tourism | M | 0.5549 (0.0335) | 0.5803 (0.0174) | 0.7457 (0.0161) | 0.7938 (0.0465) | 0.6228 (0.0095) | 0.6564 (0.0132) | 0.6554 (0.0190) | 0.9876 (0.0936) | 0.8165 (0.0240) | 0.8057 (0.0186) | 0.8230 (0.0139) | 1.0525 (0.1068) | 0.6349 (0.0107) | 0.6108 (0.0171) | 0.6114 (0.0065) | 1.1097 (0.0244) | 0.4376 (-) | 0.5105 (-) |
| Q | 0.6349 (0.0128) | 0.6428 (0.0142) | 0.7976 (0.0227) | 0.7594 (0.0459) | 0.5982 (0.0173) | 0.6236 (0.0327) | 0.7131 (0.0241) | 0.8791 (0.0860) | 0.8221 (0.0577) | 0.8556 (0.0523) | 0.8282 (0.0282) | 0.8397 (0.0431) | 0.5811 (0.0221) | 0.6212 (0.0084) | 0.7454 (0.0122) | 0.8607 (0.0384) | 0.6232 (-) | 0.6894 (-) | |
| Y | 0.8995 (0.0198) | 0.8550 (0.0295) | 0.9295 (0.0254) | 0.9524 (0.0542) | 0.8570 (0.0254) | 0.9025 (0.0279) | 0.8281 (0.0241) | 0.8323 (0.0233) | 0.7907 (0.0260) | 0.8570 (0.0377) | 0.9087 (0.0537) | 0.8869 (0.0269) | 0.8973 (0.0221) | 0.8785 (0.0221) | 0.9466 (0.0328) | 0.8508 (0.0261) | 0.8804 (-) | 1.0160 (-) |
Table I.17. Empirical evaluation of point forecasts. Mean absolute scaled error (MASE, Eq. (I.(2)) with the Naive baseline, averaged over 5 runs. Lower measurements are preferred; values below 1 beat the Naive. The methods without standard deviation have deterministic solutions. For each encoder we compare four forecasting strategies: our Continuous Horizon (CH), multi-step (MS), direct (DIR), and recurrent (REC).
The mean absolute scaled error [32] scales the absolute error by that of the Naive forecast , which repeats the last observation at the forecast creation date over the whole horizon:
We scaled by the Naive error in the same forecast windows, as in the relAv-gRMSE, rather than by the seasonal error in-sample Naive in M4’s definition.
Universal Periodic Operator Approximation Theorem
Theorem 1 guarantees the approximation only in the training interval . For periodic targets, a period that fits within the training interval is sufficient to extend the guaranty to all horizons .
Theorem 3 (Universal Periodic Operator Approximation Theorem). Assume the conditions of Theorem 1. Additionally, assume that for every history the forecast function extends to all and is periodic with a period ,
Then for any accuracy , there exists a neural network approximation
such that, for every integer ,
Proof. The proof has three steps.
One period approximation. By Theorem 1, there exist neural basis functions and coefficient mappings such that satisfies for all . Because , the period is within the training interval and the integrand is nonnegative, so .
Periodic extension. For , , so equals in the first period and .
In every period. Both and are -periodic in . The substitution with gives for every .
References
- [1]Acosta, J., et al., 2026. Effective sample size for functional spatial data. arXiv preprint arXiv:2601.20812 .
- [2]Ansari, A.F., Stella, L., Turkmen, C., Zhang, X., Mercado, P., Shen, H., Schuur, O., Rangapuram, S.S., Arango, S.P., Kapoor, S., Zschiegner, J., Maddix, D.C., Mahoney, M.W., Torkkola, K., Wilson, A.G., Bohlke-Schneider, M., Wang, Y., 2024. Chronos: Learning the language of time series. arXiv:2403.07815.arxiv.org/abs/2403.07815
- [3]Athanasopoulos, G., Hyndman, R.J., Song, H., Wu, D.C., 2011. The Tourism forecasting competition. International Journal of Forecasting 27, 822–844. URL: https://www.sciencedirect.com/science/article/pii/S0169207010000107X, doi:https://doi.org/10.1016/j.ijforecast.2010.04.009. special Section 1: Forecasting with Artificial Neural Networks and Computational Intelligence. Special Section 2: Tourism Forecasting.
- [4]Bao, Y., Xiong, T., Hu, Z., 2014. Multi-step-ahead time series prediction using multiple-output support vector regression. Neurocomputing 129, 482–493. URL: https://www.sciencedirect.com/science/ article/pii/S092523121300917X, doi:https://doi.org/10.1016/j.neucom.2013.09.010.
- [5]Ben Taieb, S., Atiya, A.F., 2015. A bias and variance analysis for multistep-ahead time series forecasting. IEEE transactions on neural networks and learning systems 27, 62–76. URL: https://ieeexplore.ieee.org/document/7064712.DOI
- [6]Ben Taieb, S., Bontempi, G., Atiya, A.F., Sorjamaa, A., 2012. A review and comparison of strategies for multi-step ahead time series forecasting based on the NN5 forecasting competition. Expert Systems with Applications 39, 94–103. URL: https://arxiv.org/abs/1108.3259.
- [7]Ben Taieb, S., Sorjamaa, A., Bontempi, G., 2010. Multiple-output modeling for multi-step-ahead time series forecasting. Neurocomputing 73, 1950–1957. URL: https://www.sciencedirect.com/science/article/pii/S0925231210001013, doi:https://doi.org/10.1016/j.neucom.2009.11.030. subspace Learning / Selected papers from the European Symposium on Time Series Prediction.
- [8]de Boor, C., 1978. A Practical Guide to Splines. volume 27 of Applied Mathematical Sciences. Springer-Verlag, New York.
- [9]Box, G.E., Jenkins, G.M., Reinsel, G.C., Ljung, G.M., 2015. Time series analysis: forecasting and control. John Wiley & Sons.
- [10]Challu, C., Olivares, K.G., Oreshkin, B.N., Ramirez, F.G., Canseco, M.M., Dubrawski, A., 2023. N-HiTS: Neural hierarchical interpolation for time series forecasting, in: AAAI, pp. –. URL: https://arxiv.org/abs/2201.12886.
- [11]Chang, S., Zhang, Y., Han, W., Yu, M., Guo, X., Tan, W., et al., 2017. Dilated recurrent neural networks, in: 31st Conference on Neural Information Processing Systems, pp. –.
- [12]Chapados, N., Bengio, Y., 2007. Augmented functional time series representation and forecasting with gaussian processes, in: Platt, J., Koller, D., Singer, Y., Roweis, S. (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc.. pp. –. URL: https://proceedings.neurips.cc/paper_files/paper/2007/ file/81e74d678581a3bb7a720b019f4f1a93-Paper.pdf.
- [13]Chen, R.T.Q., Rubanova, Y., Bettencourt, J., Duvenaud, D., 2018. Neural ordinary differential equations, in: Advances in Neural Information Processing Systems, pp. –.arxiv.org/abs/1806.07366
- [14]Chen, T., Chen, H., 1995. Universal approximation theorem for operators to neural networks. IEEE Transactions on Neural Networks 6, 911–917. doi:10.1109/72.392253.
- [15]Chevillon, G., 2007. Direct multi-step estimation and forecasting. Journal of Economic Surveys 21, 746–785. doi:https://doi.org/10.1111/j.1467-6419.2007.00518.x.
- [16]Cho, K., van Merrienboer, B., Gülçehre, Ç., Bougares, F., Schwenk, H., Bengio, Y., 2014. Learning phrase representations using RNN encoder-decoder for statistical machine translation. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) abs/1406.1078, 1724–1734. URL: http://arxiv.org/abs/1406.1078, arXiv:1406.1078.
- [17]Darlow, L., Deng, Q., Hassan, A., Asenov, M., Singh, R., Joosen, A., Barker, A., Storkey, A., 2024. Dam: Towards a foundation model for forecasting, in: International Conference on Learning Representations (ICLR), pp. –.arxiv.org/abs/2407.17880
- [18]Das, A., Kong, W., Sen, R., Zhou, Y., 2024. A decoder-only foundation model for time-series forecasting. URL: https://arxiv.org/abs/2310.10688, arXiv:2310.10688.
- [20]Garza, A., Mergenthaler-Canseco, M., 2023. Timegpt. URL: https://arxiv.org/abs/2310.03589, arXiv:2310.03589.
- [21]Griffith, D.A., 2005. Effective geographic sample size in the presence of spatial autocorrelation. Annals of the Association of American Geographers 95, 740–760.DOI
- [22]Gu, A., Goel, K., Ré, C., 2022. Efficiently modeling long sequences with structured state spaces, in: International Conference on Learning Representations, pp. –. URL: https://arxiv.org/abs/2111.00396.
- [23]Hersbach, H., 2000. Decomposition of the continuous ranked probability score for ensemble prediction systems. Weather and Forecasting 15, 559–570.doi.org/10.1175/1520-0434(2000)015<0559:dotcrp>2.0.co;2
- [24]Hochreiter, S., Schmidhuber, J., 1997. Long short-term memory. Neural Computation 9, 1735–1780.
- [25]Holt, C.C., 1957. Forecasting seasonals and trends by exponentially weighted moving averages. (O.N.R. Memorandum No. 52) URL: https://doi.org/10.1016/j.ijforecast.2003.09.015.
- [26]Hoo, S.B., Müller, S., Salinas, D., Hutter, F., 2024. The tabular foundation model tabPFN outperforms specialized time series forecasting models based on simple features, in: NeurIPS 2024 Third Table Representation Learning Workshop, pp. –. URL: https://openreview.net/forum?id=HO2X7R03OC.
- [27]Hornik, K., 1991. Approximation capabilities of multilayer feed-forward networks. Neural Networks 4, 251–257. URL: https://www.sciencedirect.com/science/article/pii/089360809190009T, doi:https://doi.org/10.1016/0893-6080(91)90009-T.
- [28]Huber, P.J., 1967. The behavior of maximum likelihood estimates under nonstandard conditions, in: Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, University of California Press. pp. 221–233.
- [29]Hyndman, R.J., Athanasopoulos, G., Garza, A., Challu, C., Mergenthaler, M., Olivares, K.G., 2026. Forecasting: Principles and Practice, the Pythonic Way. OTexts, Melbourne, Australia. Available at https://otexts.com/fpppy/.
- [30]Hyndman, R.J., Billah, B., 2003. Unmasking the theta method. International Journal of Forecasting 19, 287–290. URL: https://www.sciencedirect.com/science/article/pii/S0169207001001431, doi:https://doi.org/10.1016/S0169-2070(01)00143-1.
- [31]Hyndman, R.J., Khandakar, Y., 2008. Automatic time series forecasting: The forecast package for r. Journal of Statistical Software, Articles 27, 1–22. URL: https://www.jstatsoft.org/v027/i03, doi:10.18637/jss.v027.i03.DOI
- [32]Hyndman, R.J., Koehler, A.B., 2006. Another look at measures of forecast accuracy. International Journal of Forecasting 22, 679 – 688. URL: http://www.sciencedirect.com/science/article/pii/S0169207006000239, doi:https://doi.org/10.1016/j.ijforecast.2006.03.001.
- [33]Hyndman, R.J., Shahid Ullah, M., 2007. Robust forecasting of mortality and fertility rates: A functional data approach. Computational Statistics & Data Analysis 51, 4942–4956. URL: https://www.sciencedirect.com/science/article/pii/S0167947306002453, doi:https://doi.org/10.1016/j.csda.2006.07.028.
- [34]Hyndman, R.J., Shang, H.L., 2009. Forecasting functional time series. Journal of the Korean Statistical Society 38, 199–211. URL: https://www.sciencedirect.com/science/article/pii/S1226319209000398, doi:https://doi.org/10.1016/j.jkss.2009.06.002.
- [35]Januschowski, T., Wang, Y., Torkkola, K., Erkkilä, T., Hasson, H., Gasthaus, J., 2022. Forecasting with trees. International Journal of Forecasting 38, 1473–1481.DOI
- [36]Kidger, P., Morrill, J., Foster, J., Lyons, T., 2020. Neural controlled differential equations for irregular time series, in: Advances in Neural Information Processing Systems, pp. 6696–6707. URL: https://arxiv.org/abs/2005.08926.
- [38]Li, Z., Kovachki, N., Azizzadenesheli, K., Liu, B., Bhattacharya, K., Stuart, A., Anandkumar, A., 2021. Fourier neural operator for parametric partial differential equations. URL: https://arxiv.org/abs/2010.08895, arXiv:2010.08895.
- [39]Lu, L., Jin, P., Pang, G., Zhang, Z., Karniadakis, G.E., 2021. Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators. Nature Machine Intelligence 3, 218–229. doi:10.1038/s42256-021-00302-5.DOI
- [40]Makridakis, S., Andersen, A., Carbone, R., Fildes, R., Hibon, M., Lewandowski, R., Newton, J., Parzen, E., Winkler, R., 1982. The accuracy of extrapolation (time series) methods: Results of a forecasting competition. Journal of Forecasting 1, 111–153. doi:https://doi.org/10.1002/for.3980010202.
- [41]Makridakis, S., Hibon, M., 2000. The M3-competition: results, conclusions and implications. International Journal of Forecasting 16, 451–476. URL: https://www.sciencedirect.com/science/article/pii/S0169207000000571, doi:https://doi.org/10.1016/S0169-2070(00)00057-1. the M3- Competition.
- [42]Makridakis, S., Spiliotis, E., Assimakopoulos, V., 2020. The M4 competition: 100,000 time series and 61 forecasting methods. International Journal of Forecasting 36, 54–74. URL: https://www.sciencedirect.com/science/article/pii/S0169207019301128, doi:https://doi.org/10.1016/j.ijforecast.2019.04.014. m4 Competition.
- [44]Nie, Y., Nguyen, N.H., Sinthong, P., Kalagnanam, J., 2023. A time series is worth 64 words: Long-term forecasting with transformers, in: International Conference on Learning Representations, pp. –.arxiv.org/abs/2211.14730
- [45]Olivares, K.G., Challu, C., Marcjasz, G., Weron, R., Dubrawski, A., 2023a. Neural basis expansion analysis with exogenous variables: Forecasting electricity prices with NBEATSx. International Journal of Forecasting 39, 901–921.
- [46]Olivares, K.G., Meetei, N., Ma, R., Reddy, R., Cao, M., Dicker, L., 2023b. Probabilistic hierarchical forecasting with deep poisson mixtures. International Journal of Forecasting Preprint version available at arXiv:2110.13179. URL: https://arxiv.org/abs/2110.13179.
- [47]Olivares, K.G., Wolff, M., Konstantinova, T., Ramasubramanian, S., Oreshkin, B., Wilson, A.G., Potapczynski, A., Potosnak, W., Cao, M., Mahoney, M.W., Efimov, D., 2025. A realistic evaluation of cross-frequency transfer learning and foundation forecasting models, in: Thirty-Ninth Annual Conference on Neural Information Processing Systems NeurIPS 2025, NeurIPS 2025, San Diego, USA. pp. –. URL: https://arxiv.org/abs/2411.05852, arXiv:2411.05852.
- [48]van den Oord, A., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., Kalchbrenner, N., Senior, A.W., Kavukcuoglu, K., 2016. WaveNet: A generative model for raw audio. Computer Research Repository abs/1609.03499. URL: http://arxiv.org/abs/1609.03499, arXiv:1609.03499.
- [49]Oreshkin, B.N., Carpov, D., Chapados, N., Bengio, Y., 2020. N-beats: Neural basis expansion analysis for interpretable time series forecasting, in: ICLR, pp. –.arxiv.org/abs/1905.10437
- [50]Potosnak, W., Wolff, M., Cao, M., Ma, R., Konstantinova, T., Efimov, D., Mahoney, M.W., Oreshkin, B.N., Olivares, K.G., 2026. Forking-sequences: Statistically and computationally efficient multi-horizon forecasting with reduced volatility. Transactions on Machine Learning Research URL: https://openreview.net/forum?id=dXdycy7WCX.arxiv.org/abs/2510.04487
- [51]Ramsay, J.O., Silverman, B.W., 2005. Functional Data Analysis. Springer Series in Statistics. 2nd ed., Springer, New York. doi:10.1007/b98888.DOI
- [52]Rubanova, Y., Chen, R.T.Q., Duvenaud, D., 2019. Latent ordinary differential equations for irregularly-sampled time series, in: Advances in Neural Information Processing Systems, pp. –.
- [53]Sagawa, S., Koh, P.W., Hashimoto, T.B., Liang, P., 2020. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization, in: International Conference on Learning Representations (ICLR), pp. –.arxiv.org/abs/1911.08731
- [54]Salinas, D., Flunkert, V., Gasthaus, J., Januschowski, T., 2020. DeepAR: Probabilistic forecasting with autoregressive recurrent networks. International Journal of Forecasting 36, 1181–1191.arxiv.org/abs/1704.04110
- [55]Semenoglou, A.A., Spiliotis, E., Makridakis, S., Assimakopoulos, V., 2021. Investigating the accuracy of cross-learning time series forecasting methods. International Journal of Forecasting 37, 1072–1084. URL: https://www.sciencedirect.com/science/article/pii/S0169207020301850, doi:https://doi.org/10.1016/j.ijforecast.2020.11.009.
- [56]Shannon, C.E., 1949. Communication in the presence of noise. Proceedings of the IRE 37, 10–21. doi:10.1109/JRPROC.1949.232969.DOI
- [57]Smyl, S., 2019. A hybrid method of exponential smoothing and recurrent neural networks for time series forecasting. International Journal of Forecasting , -doi:10.1016/j.ijforecast.2019.03.017.DOI
- [58]Syntetos, M., Boylan, J., Croston, J., 2005. On the categorization of demand patterns. Journal of the Operational Research Society 56, -. doi:10.1057/palgrave.jors.2601841.DOI
- [59]Teräsvirta, T., Tjøstheim, D., Granger, C.W., 2010. Modelling Nonlinear Economic Time Series. Oxford University Press.
- [60]van der Vaart, A.W., 1998. Asymptotic Statistics. Cambridge Series in Statistical and Probabilistic Mathematics, Cambridge University Press.
- [61]Vallejos, R., Osorio, F., 2014. Effective sample size of spatial process models. Spatial Statistics 9, 66–92.DOI
- [62]Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, Ł., Polosukhin, I., 2017. Attention is all you need, in: Advances in Neural Information Processing Systems (NIPS), pp. 5998–6008.arxiv.org/abs/1706.03762
- [63]Wen, R., Torkkola, K., Narayanaswamy, B., Madeka, D., 2017. A multi-horizon quantile recurrent forecaster, in: In Proceedings of 31st Conference on Neural Information Processing Systems Time Series Workshop., pp. –. URL: https://www.amazon.science/publications/a-multi-horizon-quantile-recurrent-forecaster.arxiv.org/abs/1711.11053
- [64]Woo, G., Liu, C., Kumar, A., Xiong, C., Savarese, S., Sahoo, D., 2024. Unified training of universal time series forecasting transformers. URL: https://arxiv.org/abs/2402.02592, arXiv:2402.02592.
- [65]Xie, S.M., Pham, H., Dong, X., Du, N., Liu, H., Lu, Y., Liang, P., Le, Q.V., Ma, T., Yu, A.W., 2023. DoReMi: Optimizing data mixtures speeds up language model pretraining, in: Advances in Neural Information Processing Systems (NeurIPS), pp. –.arxiv.org/abs/2305.10429







