build training-free forecasting pipelines

Time-series forecasting suffers from poor generalization and low efficiency due to tight coupling among sequence representation, information extraction, and future projection. To address this, we propose a modular forecasting framework that decouples the pipeline into three independently optimizable stages—representation learning, information extraction, and target projection—enabling flexible, task-aware component configuration. Our approach innovatively integrates convolutional layers with a lightweight self-attention mechanism, achieving efficient local feature modeling while capturing long-range temporal dependencies. Evaluated on seven benchmark datasets, the method consistently outperforms existing state-of-the-art models in prediction accuracy, while requiring significantly fewer parameters and achieving faster training and inference speeds. This demonstrates substantial improvements in both statistical performance and computational efficiency.

This work addresses the instability in multi-step probabilistic forecasting, where erratic fluctuations in predictions can undermine downstream decision-making and system reliability. The authors propose a distribution-free probabilistic forecasting method that models the conditional quantile function using neural-network-parameterized regression splines. Their approach jointly optimizes predictive accuracy and temporal stability by incorporating an explicit stability regularizer into the training objective, which penalizes discrepancies between successive forecast updates. Crucially, the framework allows for region-specific weighting—enabling enhanced robustness in critical areas such as distribution tails or the central region. This is the first distribution-agnostic method to co-optimize stability and accuracy, and it demonstrates consistent efficacy across two datasets with markedly different statistical properties, significantly reducing prediction instability while preserving high forecast quality.

This study addresses key challenges in business and financial forecasting—namely poor reproducibility, low model transparency, and weak cross-environment consistency—by systematically evaluating Meta’s open-source Prophet framework. Under a unified experimental design, Prophet is benchmarked against multiple ARIMA variants and Random Forest models, leveraging its additive structure, standardized workflow, and open implementation. The findings demonstrate that Prophet achieves competitive predictive performance while substantially enhancing reproducibility, auditability, and engineering integration efficiency. Rather than introducing a novel algorithm, this work advocates for Prophet as a transparent, reliable, and collaboration-friendly forecasting methodology, particularly suited for high-stakes decision-making contexts where interpretability and robustness are paramount.

This work addresses the limited stability of conventional time series forecasting models, which rely solely on unidirectional inference from historical observations to future targets and neglect the structural information embedded in the unobservable trajectory beyond the prediction horizon. To overcome this limitation, we propose the KUP-BI paradigm, which for the first time treats the post-target continuation as a structured prior. Leveraging knowledge distillation, we construct a lightweight proxy of this continuation from a historical trajectory repository and introduce a feature-gated fusion module to enable bidirectional interaction between input features and the continuation proxy at the representation level. Notably, our approach requires no additional external data and integrates seamlessly with mainstream time series backbones, achieving significant state-of-the-art performance gains across six public benchmarks with minimal computational overhead.

To address performance degradation in non-stationary time series forecasting caused by distributional shift, this paper proposes a decoupled modeling framework that separates distribution correction from forecasting and introduces a bilevel optimization paradigm for joint learning. Its core innovation is Instance Normalization Flow (IN-Flow)—a reversible, bidirectional, and highly expressive temporal distribution transformation network explicitly designed for forecasting, overcoming the limitation of conventional normalizing flows restricted to generative tasks. IN-Flow integrates instance normalization layers with stacked invertible neural networks: the outer level optimizes distribution transformation, while the inner level optimizes forecasting, ensuring compatibility with arbitrary forecasting architectures and eliminating reliance on statistical assumptions. Extensive experiments on synthetic and diverse real-world datasets demonstrate significant improvements over state-of-the-art methods, strong robustness to unseen distribution shifts, and simultaneous gains in both predictive accuracy and generalization capability.

This work addresses the issue of quantile crossing and inconsistency with candlestick geometric structure in probabilistic candlestick forecasting. To resolve this, we propose the K-line–Quantile Sequential Projection (KQSP) method, a post-processing technique that enforces strict quantile monotonicity and candlestick shape constraints on probabilistic predictions from any pretrained model—without requiring architectural modifications or retraining. KQSP employs a parameter-free, training-agnostic sequential projection algorithm, uniquely eliminating both types of crossing simultaneously for the first time. Experimental results demonstrate that KQSP reduces both quantile crossing rates and candlestick crossing rates to zero across all test datasets, achieving these guarantees with significantly smaller correction magnitudes compared to existing approaches, while preserving the original forecast accuracy.

This study addresses the limitation of existing forecasting systems that rely predominantly on point predictions and thus fail to adequately characterize uncertainty for informed decision-making. To overcome this, the authors propose a hybrid framework that extends point forecasts from classical models—such as Theta, exponential smoothing, and ARIMA—into probabilistic forecasts by integrating error post-processing with model-specific, horizon-dependent uncertainty scaling. The approach calibrates forecast errors using historical simulation, conformal prediction, quantile regression, and GARCH-based methods, and systematically evaluates in-sample versus out-of-sample calibration performance. Empirical results on the M4 dataset demonstrate an average 4.6% reduction in Continuous Ranked Probability Score (CRPS). In-sample calibration consistently outperforms out-of-sample calibration, particularly over longer forecast horizons, thereby validating the effectiveness and practical utility of the proposed framework.

This work addresses key challenges in long-term time series forecasting—namely, high model dependency, difficulty in capturing the full predictive distribution, and delayed feedback—by proposing KReF, a training-free framework. KReF introduces retrieval as an inductive bias for the first time, leveraging either handcrafted statistical features or frozen random Fourier feature embeddings of the lookback window to retrieve historically similar segments and construct a local empirical distribution. Point forecasts, quantiles, and uncertainty intervals are then generated via similarity-weighted aggregation, with prediction intervals adaptively calibrated using probability integral transform. Evaluated across six benchmark datasets and four forecast horizons (12 settings total), KReF achieves the lowest Continuous Ranked Probability Score (CRPS) overall, obtains the best Interval Score at 90% coverage (IS90) in nine settings, and outperforms trainable baselines in point forecasting on two out of six datasets.

This work proposes a training-free framework that formulates time series forecasting as a planning problem, synergistically integrating the textual reasoning capabilities of large language models (LLMs) with the numerical prediction power of frozen time series foundation models (TSFMs), such as Chronos or TimesFM. The approach employs a TSFM as a trajectory simulator to generate candidate forecasts, while two role-specialized LLMs act as a policy (Ranker) and a value function (Judge), respectively. Guided by Monte Carlo Tree Search (MCTS), the method selects the optimal forecast trajectory under natural language conditions while preserving temporal structure. Experiments on the Context-is-Key and Time-MMD benchmarks demonstrate consistent and significant performance gains across diverse TSFM–LLM pairings, establishing the first training-free, cross-modal framework for text-conditioned time series forecasting.