Journal of Forecasting最新文献

Fire safety is a primary concern in safeguarding lives and property. However, it is challenging to predict fire incidents and identify potential influencing factors due to limitations of data, model accuracy and interpretability. This paper proposes a novel scheme designed to enhance predictive and explainable capabilities by integrating multi-source data, adaptive machine learning methods, and Shapley additive explanation (SHAP) tools for more effective and applicable fire safety management. The scheme shows satisfactory prediction results by leveraging the data from grid-style management systems and our proposed machine learning method with dynamic time warping distance-based time series clustering, significantly outperforming the methods merely based on time series modeling. Moreover, clustered features help to clarify the main influencing risk factors and provide clearer insights for model interpretability. With global SHAP, community clusters capturing community fire event frequency, as well as historical records on fire police rescue, smoke alarms, and fire alarms, are found to be significant risk factors among all the features over the whole communities and periods via the model interpretability analysis, implying that communities where fires used to occur frequently are more likely to occur in future, which should be highly vigilant in real fire management. With local SHAP, specific risk factors that vary across communities can be identified for any single community with a given period. We demonstrate the potential of this integrated machine learning scheme in improving the prediction accuracy and risk identification applicability of fire incidents, which contributes to more effective and customized fire safety management.

Global forecasting models (GFMs) have become essential in time series prediction, as they enable cross-learning across multiple series. Although GFMs have consistently outperformed univariate approaches, their performance decreases when applied to heterogeneous time series datasets, such as those found in economic and financial applications. Clustering techniques have been used to create homogeneous time series clusters. However, the main limitations of current clustering-based GFMs are as follows: (1) employing handcrafted features instead of deep learning and (2) there is no guarantee that the resulting clusters are optimal in terms of prediction accuracy. To address these limitations, we propose a novel deep time series clustering model that jointly optimizes clustering and forecasting accuracy. The proposed method simultaneously optimizes the reconstruction, clustering, and prediction losses to ensure clusters are optimized for accurate forecasting. In addition, it employs a neighbor-aided autoencoder to capture cluster-oriented representations, leveraging neighboring time series to improve feature learning. Furthermore, we incorporate an evolutionary learning component, which iteratively refines clusters through crossover and mutation to find optimal clusters in terms of forecasting accuracy. We evaluate our proposed method on eight publicly available datasets considering various state-of-the-art forecasting benchmarks. Results indicate that across all datasets with 2620 time series, the proposed method obtains the lowest mean symmetric mean absolute percentage error (sMAPE) of 14.90, surpassing the baseline deep clustering (15.15). It exhibits enhancements of 1.28, 0.70, and 2.29 in mean sMAPE relative to DeepAR, N-BEATS, and transformer, respectively. Furthermore, it demonstrates improvements when compared to the existing clustering-based global models. The source code of the proposed clustering method is made publicly available at https://github.com/alinowshad/Evolutionary-Neighbor-Aided-Deep-Clustering-DEEPEN.

Early studies that used rolling windows found it to be a useful forecasting technique. These studies were, by-and-large, based on pre-2000 data, which were nonstationary. Subsequent work, based on stationary data from the mid-1990s to 2020, has not been able to confirm that finding. However, this latter result may reflect the fact that there was relatively little structural instability between the mid-1990s and 2020: The data had become stationary. Following the series of shocks of the early 2020s, this is no longer the case because the shocks produced nonstationarity in the macroeconomic data, such as inflation. Consequently, rolling windows may again be a sensible way forward. The present study assesses this conjecture.

Compared to make-to-order production based on customer order, make-to-stock based on forecast can effectively reduce inventory level and production cost. However, due to high randomness of spot markets and many uncertainties in production environments, it is hard to forecast the products accurately. In this article, a hybrid model combining seasonal autoregressive integrated moving average (SARIMA) and least square support vector machines (LSSVMs) is proposed to forecast the potential demand of steel products. First, the SARIMA based on a multiobjective differential evolution with improved mutation strategies is developed to extract linear components of the potential demand. Then, a sparse strategy is designed to extract useful data and hence reduce computation complexity without loss of accuracy. Next, the LSSVMs combined with a single-objective differential evolution are adopted to extract nonlinear components of the potential demand. Finally, the experimental results on a real-world instance demonstrate the effectiveness of the proposed model and algorithm.

This study utilizes modal regression to forecast the cumulative confirmed COVID-19 cases in Canada, Japan, South Korea, and the United States. The objective is to improve the accuracy of the forecasts compared to standard mean and median regressions. To evaluate the performance of the forecasts, we conduct simulations and introduce a metric called the coverage quantile function (CQF), which is optimized using modal regression. By applying modal regression to popular time-series models for COVID-19 data, we provide empirical evidence that the forecasts generated by the modal regression outperform those produced by the mean and median regressions in terms of the CQF. This finding addresses the limitations of the mean and median regression forecasts.

This study evaluates the economic usefulness of stock market implied volatility forecasts, based on their ability to improve the short-run trading decision-making process. The current literature aligns the forecast horizon with the frequency of the trading decision in order to evaluate different forecasting frameworks. By contrast, the premise of our paper is that these should not be necessarily related, but rather the evaluation should be based on the actual needs of the end-user. Thus, we evaluate whether the multiple days ahead stock market volatility forecasts vis-à-vis the 1-day ahead forecasts can improve the 1-day ahead trading profits from VIX and the S&P500 futures. Our results suggest that indeed the 1-day ahead trading profits are significantly improved when the trading decisions are based on longer term volatility forecasts. More specifically, the highest trading gains are obtained when using the 22-day ahead forecasts. The results hold true for both VIX and S&P500 futures day-ahead trading. Although there is no theoretical background regarding the fact that forecasting and trading horizons should not be aligned, we strongly motivate this potential issue, both from the statistical and financial points of view.

We examine the contribution of a shock to climate concern to the observed outperformance of a portfolio of European green stocks relative to a brown benchmark. We show, first, that an information set given by 1-month stock return and realized volatility of each stock constituent (and their cross-sectional averages) improves the (in-sample) forecasting performance for the return series relative to the traditional market risk factors proxied by Fama–French portfolios. Moreover, the identification of the shock to climate concern occurs in two stages: First, we compute the historical decomposition based on a Panel SVAR fitted to the return and volatility of each green and brown portfolio constituent. Then, the contribution of the first common shock to the historical decomposition of returns is purged of macroeconomic forecast errors, and the residual is interpreted as the innovation to climate concern. The empirical evidence is robust to a number of different selections of stocks entering the green and brown portfolio.