Journal of Forecasting最新文献

Efficient prediction of air pollutant concentration is of great significance to air pollution prevention, human health protection, and cleaner production. Previous air quality studies mainly focused on point-based and interval-based forecasts, and a potential problem with these methods is that some important information may be lost. Therefore, this paper puts forward a novel ternary interval decomposition ensemble paradigm for air pollutant concentration forecasting, which is capable of capturing the daily minimum, daily average, and daily maximum of air pollutant concentration concurrently. In this paradigm, the ternary interval-valued air pollutant concentration time series (TIAPCTS) is innovatively constructed. Multivariate empirical mode decomposition is first applied to decompose the TIAPCTS into multiple ternary intrinsic mode functions (TIMFs) and one ternary residue (TR). Then, the lower, middle, and upper bounds of each TIMF and TR are simultaneously fitted and predicted by the three-output multivariate relevance vector machine. To obtain better final outputs, an optimal variable weight ensemble approach is suggested to integrate the forecasting results of TIMFs and TR. The novelty of this study comes from the ternary interval forecasting perspective, multivariate modeling techniques, and weighted ensemble strategy, which not only fully take possible associations among the lower, middle, and upper bounds into account but also improve the modeling efficiency while generating multiple correlated outputs. The proposed paradigm is justified with six kinds of real-world air pollutant concentration data from Beijing and Shanghai, China, indicating it is a promising alternative for air pollution concentration analysis and forecast.

For many years, time-series data analysis and prediction has been extensively studied, developed, and showed great potential in dealing with multiple real-world problems, including long-term financial forecasting, and early extreme weather condition forecasting. The rapid evolution of deep neural networks has significantly advanced the field of time-series forecasting, offering powerful tools for analyzing and predicting complex temporal patterns. Transformers, in particular, have gained widespread adoption for their ability to model long-range dependencies through the multiheaded self-attention (MHA) mechanism, making them effective for both univariate and multivariate time-series (MTS) tasks. However, when applied to complex MTS data, most traditional transformer-based techniques face notable challenges. The intricate temporal dependencies and cross-variable relationships in MTS are often too complex for standard self-attention mechanisms to fully capture. Thus, it leads to limitations in forecasting performance. Moreover, existing transformer-based approaches frequently neglect the critical cross-variable interactions that are essential for accurate multivariate forecasting. Moreover, previous techniques are also limited in comprehensively evaluating the temporal patterns at the series level. These temporal patterns are crucial for understanding how historical trends influence future predictions. To address these limitations, we propose a novel TAT4MTS model, which is a trend-aware transformer-based architecture that effectively captures intricate cross-dependent patterns and produces enhanced series-level representations, enabling significant improvements in long-term forecasting accuracy. The trend-aware projection mechanism, which is applied in our model, can assist in effectively discovering and capturing intricate cross-dependent patterns in MTS data. As a result, it generates enhanced aggregated series-level representations of input sequences, thereby improving its ability to model complex temporal relationships. Extensive empirical studies within real-world multivariate weather datasets validate the effectiveness as well as the outperformance of our proposed model, comparing with previous transformer-based forecasting baselines.

This study investigates the relationship between cyber risk and systemic risk using firm-level data from 2006 to 2018. We employ machine learning techniques to develop a predictive model for cyber risk and assess its impact on asset correlation, a proxy for systemic risk. Our analysis reveals that higher cyber risk is significantly associated with increased systemic risk. The results are robust across various checks, including the exclusion of financial firms and the financial crisis period. Furthermore, we find that the cyber-related component of systemic risk has a substantial impact on future stock returns, indicating a significant risk premium. These findings highlight the importance of integrating cyber risk into traditional risk management and asset pricing models, providing valuable insights for investors and policymakers.

The aim of this study is to identify the characteristics of policyholders that may indicate a risk of cancellation in the French insurance sector over the period 2013–2016. Customer churn predictions are provided by using traditional data mining methods (Tobit model), machine learning methods (XGBoost algorithm), and a hybrid of machine learning and data mining methods (Grabit model). Results show that the Grabit model outperforms the Tobit model and XGBoost algorithm according to selected performance metrics. The study revealed that the characteristics of clients (age, marital status) and the characteristics of the policies (type, number of policies, payment mode, online channel, timing of policy cancellations) significantly influence client departure. They allow for better decision-making and the implementation of relevant marketing strategies.

To mitigate the risks posed by the forgery of SIM card information, couriers must conduct face-to-face real-name verification with consumers during door-to-door delivery. However, due to various factors, the failure rate of the first delivery handover is high, leading to the need for one or even multiple appointment deliveries. Moreover, the appointment delivery time is often broad, making it difficult to provide precise guidance for couriers' delivery plans, thereby affecting overall delivery efficiency. To address the above issues, this paper explores the problem of predicting the delivery time of SIM cards in the case of first delivery failure. First, this paper comprehensively considers factors that affect the delivery time, including distance, weather, day of the week, and consumer features. These factors are incorporated into the prediction model to make the model more applicable to real-world scenarios. Second, this paper constructs the NSGA-III-XGBoost model. This model uses XGBoost (eXtreme Gradient Boosting) as the base model and optimizes the hyperparameters of XGBoost by using NSGA-III (Non-dominated Sorting Genetic Algorithm III). Experimental results show that the proposed model outperforms several benchmark models in terms of prediction accuracy and stability. Finally, this paper uses the SHAP (Shapley Additive exPlanations) method to explain the prediction results of the NSGA-III-XGBoost model and deeply explores the impact of various features on delivery time prediction. Based on the experimental results, this paper summarizes the research findings and provides references for model construction and experimental design. By predicting the delivery time of SIM cards in the case of first delivery failure, this paper provides couriers with a reference for the final delivery time, helping them make reasonable arrangements within the time slots scheduled with consumers and thereby enhancing delivery efficiency.

The current research presents a novel approach that integrates the first-moment (mean) and second-moment (variance) components of stock price dynamics to forecast future price trends. Employing a combination of statistical and deep learning models, the study aims to predict both the mean and variance of stock price movements for select pharmaceutical companies in India based on their market capitalization. The forecasts are then utilized to assess the effectiveness of the Bollinger Band (BB) trading strategy in terms of hit ratio and average returns per trade. The study covers both pre- and post-COVID periods. The results indicate that the integrated mean and volatility model employed in this study outperforms the stand-alone mean and volatility models when back-tested with BB trading strategies, leading to higher returns. Moreover, when combined with a volatility model, the integrated deep learning model consistently demonstrates superior performance compared with the standalone mean or volatility model. The integrated model has yielded significantly higher annualized average returns (> 200%) than the returns generated based on technical indicators, as suggested by existing studies. These findings have significant practical implications, providing investors and traders with an advanced alternative to conventional trading methods.

The extant literature on forecasting interval-valued crude oil prices has predominantly focused on estimating the conditional mean, while little attention has been paid to the conditional distribution, which may offer more insightful information. This paper proposes a novel model free interval-based Treeffuser approach to forecast the conditional distribution of interval-valued crude oil prices. This approach involves estimating the score function in the inverse stochastic differential equation using gradient boosting trees and generating samples through a diffusion process. Empirical results demonstrate that the proposed approach outperforms classical interval-based machine learning methods, particularly during extreme events. The superior performance is robust to various forecast horizons and estimation periods. Furthermore, we propose an interval-based trading strategy that can effectively mitigate volatility and boost returns. Notably, our approach captures the significant impact of extreme events on minimum prices, revealing a dynamic pattern where the range of prices initially widens, and later the distribution of minimum prices flattens out.

This paper proposes a sparse scaled principal component analysis (PCA) to forecast stock returns. The sparse scaled PCA is a modification to the common PCA by first selecting the important predictors and then scaling the selected predictors according to their predictive power on the target to be forecasted. An advantage of the proposed sparse scaled PCA is that it takes the benefits of both scaled PCA and supervised PCA. An empirical study is conducted on the US stock market to evaluate its empirical performance, and the results confirm the superiority of sparse scaled PCA over a variety of dimension-reduction techniques, including PCA, PLS, scaled PCA, and supervised PCA in both in-sample and out-of-sample forecasting. Economic value analysis shows that the outperformance can yield economic utility gains at a reasonable transaction cost.