Journal of Forecasting最新文献

Repurchase prediction is a vital aspect of marketing strategy and a complex decision-making task, especially in the airline industry, where data are uncertain, incomplete, and ambiguous. To address this, this study proposes a novel multi-classifier evidence ensemble algorithm that integrates evidence theory with machine learning to predict travelers' repurchase behavior. The model was trained using 29 behavioral features derived from a low-cost Chinese airline. Empirical results show that the proposed algorithm outperforms traditional models in terms of the accuracy, the precision, the recall, the F1-score, and the AUC. Specifically, it achieved over 80% accuracy and precision in binary classification tasks. Ablation experiments using four classifier combinations at different sampling rates (30%, 50%, and 70%) further validated the robustness and effectiveness of the framework. The results suggest that the proposed ensemble framework outperforms traditional prediction models in terms of overall predictive performance for analyzing airline passenger behavior in real-world settings.

This paper introduced HyperVIX, a novel hybrid framework that integrates ARIMA modeling, LSTM neural networks, and Gray Wolf Optimizer (GWO) to forecast the Chicago Board Options Exchange (CBOE) Volatility Index (VIX). Using a multilayered approach, HyperVIX first employs ARIMA to capture linear time series patterns, followed by LSTM networks that model the residuals to identify complex nonlinear relationships. The GWO algorithm optimizes the LSTM hyperparameters, enhancing the framework's ability to capture Volatility Index (VIX)'s intricate dynamics. Empirical analysis demonstrates that HyperVIX significantly outperforms both traditional and contemporary financial forecasting models in terms of accuracy and robustness. Compared to single models, HyperVIX achieves approximately 15%, 12%, and 10% improvements in root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) metrics respectively, with the R² value increasing by about 5%. Notably, the model exhibits exceptional performance during extreme market volatility periods, making it particularly valuable for risk management applications. This research contributes to the literature by providing an innovative and effective method for VIX forecasting while offering valuable insights for financial market volatility analysis and investment strategy optimization.

There are 21,999 China utility model patents with existing decisions of invalidation reexamination from 2000 to 2021 to explore application of support vector machine (SVM) with Gaussian radial basis function (RBF) kernel. This study identified significant patent indicators using analysis of variance (ANOVA), Kruskal–Wallis test, and Jonckheere–Terpstra ordered-alternatives test and employed SVM incorporating significant patent indicators to forecast decision of invalidation reexamination with highest accuracy for patents with fully invalid claims. The study confirmed SVM with RBF to forecast patent sustainability and providing support for due diligence in mergers and acquisitions and litigation strategies.

This study adopts a component-driven approach to improve FX volatility and value-at-risk (VaR) forecasts, with a focus on two types of leading indicators: currency indexes and sovereign spreads. Specifically, we explore the significance of the US dollar index, RMB index, and China–US 10-year sovereign bond yield spread, as long-term volatility components in GARCH-MIDAS models for the USDCNH exchange rate. The investigation reveals that these explanatory variables have a substantial influence on the market's volatility. In terms of the enhanced prediction of volatility and VaR, our analysis presents empirical evidence for the forecasting superiority of the GARCH-MIDAS models that fully exploit the aforementioned variables and their combinations. Improving upon the traditional method, the optimal GARCH-MIDAS specification is comparable to or even outperforms the intraday high-frequency realized volatility model. Our research contributes to a deeper understanding of the influential factors behind USDCNH fluctuations and advances an effective method to accurately forecast volatility and VaR from component-driven perspectives.

Accurate precipitation prediction is vital for effective water resource management, agricultural planning, and natural disaster mitigation. Traditional forecasting methods often encounter difficulties due to the nonlinearity, complex seasonality, and noise inherent in meteorological data. This paper introduces a novel methodology that combines the FB–Prophet algorithm, designed by Facebook for identifying trends and seasonal patterns, with a fuzzy clustering algorithm. This integration aims to refine a crucial aspect of the FB–Prophet framework: the identification and incorporation of special events, specifically holidays, which play a significant role in the predictive modeling process. This approach ensures that holidays are effectively integrated into forecasts, enhancing the model's overall accuracy and reliability. Additionally, the proposed model is compared to several widely used algorithms in recent studies in terms of accuracy, employing nonparametric tests for a robust evaluation. Empirical results demonstrate a significant improvement in forecast accuracy over traditional methods.

This study examines the predictive performance of Feature Attention (FA), Temporal Attention (TA), and Feature and Temporal Attention (FATA) within Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Transformer architectures using price data from four Chinese carbon markets (CEA, BEA, GDEA, and HBEA). Drawing on multiple forecasting accuracy measures and significance testing, the results show that attention mechanisms can enhance forecasting accuracy in certain market-model combinations, but their effectiveness critically depends on the alignment among market conditions, model architectures, and attention mechanisms. In markets with high average prices and volatility, FA achieves the best performance with GRU and LSTM; in lower price, moderately volatile markets, TA combined with Transformer is more effective; and in the high-price, high-volatility CEA market, FATA shows promise when paired with Transformer, but lacks robustness across markets. These findings highlight a pronounced compatibility pattern among market conditions, model architectures, and attention mechanisms, suggesting that the deployment of attention mechanisms in carbon price forecasting should be tailored to specific market conditions and model structures rather than applied universally.

Accurate volatility prediction is essential for guiding investor decision-making, assessing financial stability, and managing risk. The efficacy of volatility prediction is an important issue that hinges on selecting relevant factors and applying robust analytical tools. Therefore, we propose a novel decomposition-learning method that integrates deep-learning techniques for volatility prediction. Specifically, the study employs convolutional neural networks (CNN) and long short-term memory (LSTM) networks to capture the nonlinear features in time series data. To enhance the model's predictive capabilities, we introduce singular spectrum analysis (SSA) and develop a feature contribution evaluation algorithm to identify and filter out the factors exerting the greatest influence. Building on this foundation, we construct the SSA-CNN-LSTM model that supports dual volatility prediction and evaluates each feature's contribution. We design and implement the framework and algorithms for this new approach, applying it to volatility prediction for major global stock indices. The results show that: (1) the trend, cycle, and perturbation components extracted from realized volatility outperform external factors in prediction; and (2) eliminating the features with the lowest contribution significantly enhances the model's predictive performance, thus providing financial markets with a more accurate volatility prediction tool.

Considering that, in the recent past, several studies have reasserted that human judgment is still valuable for forecasting in supply chains, this paper proposes a demand forecasting decision model (DFDM), which mathematically integrates expert judgment estimates with forecasts generated from time series and machine-learning models. An interval 3-point elicitation procedure has been used to effectively obtain estimates from experts, and further, a mathematical bias adjustment mechanism is used to detect and eliminate any systematic bias in experts' forecasts. The real-life data from manufacturing and retail firms are collected to test the proposed model. The independent variables in the data are taken as interval data series rather than crisp values to capture the uncertainty in the variables and use them for forecasting models. Error metrics to measure bias (mean error), accuracy (mean absolute error), and variance (mean squared error) of forecasts were used to evaluate the performance of the proposed DFDM. The forecasts from the proposed model were found to be significantly better than those from popular forecasting methods in practice. Finally, using temporal disaggregation, an extension to the proposed models is presented to generate very short-term forecasts to help managers make better short-term operational decisions.