Assessing and improving judgemental probability forecasts

This paper discusses three characteristics of subjective probabilities, their consistency, coherence and calibration and shows how they are inter-related. We review research that has investigated these characteristics and evaluate methods that have been developed for the improvement of judgemental probabilistic forecasting.     

The effect of the future on today's decisions

What is the effect of the future on today's decisions? The future plays a part in all of our decisions whether we utilize formal forecasting techniques or not. Some of the uncertainty of the future can be reduced by applying one or more of the techniques of trend extrapolation, subjective opinion of experts, and construction of scenarios. The results, to be useful to decision makers, must be pertinent, credible and capable of realization.     

Planning models in the UK: A survey

This paper presents the results of a survey on the use of computer-based models in the broad areas of planning and strategy formulation in UK companies. This survey updates and complements other recent surveys, which exclusively considered the use of corporate models, and shows a continued increase in the numbers of these models in use. In particular, models in the following broad categories were examined: forecasting, marketing, personnel, production and models of ‘one-off’ situations, as well as corporate and financial models. For such models the paper includes details of: frequencies of occurrence and relationship to company size; professional roles of the model-builders and users respectively; frequencies of use. In these and more subjective ways, the survey attempts to assess the place and value of computers in planning.     

具有最优学习率的RBF神经网络及其应用

传统固定学习率的RBF神经网络在金融时间序列预测方面已经有比较成功的应用,但网络学习率的选择问题却给传统RBF神经网络的使用带来了不便.利用梯度下降法及优化方法推导出了RBF神经网络的动态最优学习率并将其应用于网络学习算法,具有最优学习率的RBF神经网络能够在保证网络稳定学习的同时兼顾网络的收敛速度.为了检验具有动态最优学习率的RBF神经网络的预测效果,对沪深300指数波动率进行了预测实验.实验结果表明,具有动态最优学习率的RBF神经网络比传统的固定学习率的RBF神经网络有着更快的收敛速度,同时也避免了人为选定学习率的不便.     

基于向量夹角余弦的组合预测模型的性质研究

基于向量夹角余弦的组合预测是一种相关性的组合预测模型,它是研究组合预测方法的一个新途经.针对基于向量夹角余弦准则下组合预测模型,研究它的基本结构特征.首先提出新的优性组合预测、预测方法优超、冗余度等概念.然后探讨了非劣性组合预测、优性组合预测以及冗余预测方法的存在性,并给出冗余信息的判定定理.最后进行实例分析,表明该方法有较大的实际应用价值.     

基于修剪平均的神经网络集成时序预测方法

由于复杂时序存在结构性断点和异常值等问题，往往导致预测模型训练效果不佳，并可能出现极端预测值的情况。为此，本文提出了基于修剪平均的神经网络集成预测方法。该方法首先从训练数据中生成多组训练集，然后分别训练多个神经网络预测模型，最后将多个神经网络的预测结果使用修剪平均策略进行集成。相较于简单平均策略而言，修剪平均策略不容易受到极值的影响，能够使集成模型获得鲁棒性强的预测效果。在实证研究中，本文构造了两种神经网络集成预测模型，分别为基于修剪平均的自举神经网络集成模型(Trimmed Average based Bootstrap Neural Network Ensemble, TA-BNNE)和基于修剪平均的蒙特卡洛神经网络集成模型(Trimmed Average based Monte Carlo Neural Network Ensemble, TA-MCNNE)，并采用这两种模型对NN3竞赛数据集进行预测，结果表明在常规和复杂数据集上，修剪平均策略比简单平均策略具有更好的预测精度。此外，本文将所提出的集成模型与NN3的前十名模型进行比较，发现两种模型在全部数据集上均超过了第6名，在复杂数据集上的表现均超过了第1名，进一步验证本文所提方法的有效性。     

多因素数据重心预测方法及应用研究

本文基于数据重心概念,通过大量的研究和推导,提出数据重心参数估计理论,并利用数据重心法估计多项式回归预测模型的参数,应用于我国钢材消费量的预测。从应用的结果看,增加了我国目前钢材预测的方法。本方法能最大限度地平滑预测模型的误差,而且应用条件比最小二乘法宽松;也不会因为个别残差较大的异常点而对预测结果产生不稳定性,从而提高了拟合和预测的稳健度,计算更简捷。     

Calendar of events

Abstract. Counterbalancing is a new method of forecasting that reduces the systematic component of forecasting error. A graphical interpretation of the method is presented. This intuitive approach reveals the need for variable as opposed to fixed equal weights. The method is expanded to counterbalancing with variable weights, resulting in further reductions in forecasting error. Important applications include (1) power system hourly load forecasting for economic dispatch, (2) information feed forward in continuous process control, and (3) forecasting for scheduling, just in time manufacturing, sales, and distribution requirements planning in global logistics.     

An appreciation of alternative sales forecasting models: Recent techniques based on historical data

The subject of statistical sales forecasting has recently been brought to prominence as a major area of management decision-making by the growth of a substantive literature and the establishment of several research groups to investigate the building of forecasting models. Whilst a lot of attention has been focused on the relatively newer disciplines of technological forecasting and multiple-equation model building of macro-economic systems, statistical sales forecasting methods have also been subject to considerable development and a number of radically new techniques have emerged. The authors believe that it is timely to review the current ‘state of the art’ of sales forecasting methodology. This study examines a wide range of models in use although it is not intended as a comprehensive guide.     

基于季节调整和Holt-Winters的月度负荷预测方法

针对负荷序列中异常数据会导致模型误设或参数估计发生偏差的问题,提出利用季节调整方法,先对原始负荷序列进行季节调整,获得消除离群值、节假日影响的季节调整后序列和季节成分序列;然后用改进的Holt-Winters方法对季节调整后成分进行预测,用虚拟回归方法预测季节成分序列;最后对各成分预测结果重构得到最终预测结果的月度负荷预测方法。通过实例检验,提出的方法能明显提高预测精度,预测效果要优于季节性Holt-Winters、SARIMA、神经网络、支持向量机等模型。     

基于漂移度的组合预测方法研究

由于不同的预测方法能够提供不同的有用信息,其预测精度往往也存在差异,为了分散预测的风险,采用组合预测方法。本文首先提出相容方法集和互补模型集,然后在对不同单一预测模型的漂移性和互补性研究的基础上提出了基于漂移度的组合预测模型,为组合预测模型研究提供一种新的思路。最后通过实例来说明基于漂移度的组合预测模型能够提高样本期预测精度和外推预测精度及实际应用的有效性。     

SELECTING PARAMETERS FOR SHORT-TERM FORECASTING TECHNIQUES

The objective of this paper is to discover which of three forecasting modes used to select parameters for four short-term forecasting techniques minimizes errors. The study also examines whether the amount of historical data used to find parameters contributes to forecasting success. The results show the traditional one-ahead search routine works well in some, but not all, forecasting situations. Also, forecasting errors appear to decline when more historical data are included in the parameter search.     

二重趋势性季节型电力负荷预测组合灰色神经网络模型

对于具有增长和波动二重趋势性的季节型电力负荷,首次提出了季节型负荷预测的组合优化灰色神经网络模型,研究了同时考虑两种(非线性)趋势的复杂季节型负荷预测问题,说明了此优化模型分别优于两种单一发展趋势负荷预测模型,给出了电力负荷预测的应用实例,为季节型电力负荷预测提供了一种新的、有效的方法。     

基于期望效用-熵模型的基金评级方法及其在中国基金评级中的应用

基金评级对于投资者来说具有重要的参考价值,研究合适的基金评级方法非常必要。本文针对晨星评级对风险调整和预测能力不足的特征,研究应用期望效用-熵（EU-E）模型基金评级方法对我国开放式基金进行评级的预测能力;并以Sharpe指数、Jensen、Fama-French三因素和Carhart四因素α作为业绩指标,利用固定效应面板数据回归模型对期望效用-熵模型和随机效应面板数据回归模型对晨星基金评级的预测能力进行比较分析。采用样本期由2011年2月到2016年6月的261只基金为研究样本进行评级;研究结果表明,基于期望效用-熵平衡系数λ=0.25和0.75时,EU-E模型基金评级方法评级具有良好的预测能力,而晨星评级预测能力较弱。特别地,λ=0.25和0.75时,EU-E模型评级的五星级基金业绩优于晨星评级对应的基金业绩,而且相比于晨星评级可以更好地区分不同星级基金的业绩。另外,研究结论对于短期、中期和长期的样本都是稳健的。     

基于滚动时间窗的碳市场价格分解集成预测研究

提高碳市场价格预测准确性对于交易风险监测以及碳市场平稳发展具有重要价值。针对复杂的、非线性碳市场价格数据的短期预测误差偏大、分解过程易产生数据泄露问题，提出了基于滚动时间窗的SSA-SVR分解集成预测框架。首先，选取时间窗数据，继而借助奇异谱分析将时间窗内碳价序列分解重构为高、低频序列；然后，使用支持向量回归方法对高、低频序列分别进行预测；最后，加和集成预测结果，得到下一时刻的碳市场价格预测值。通过不断更新时间窗的数据内容，动态执行“分解-预测-集成”过程，实现碳市场价格的实时预测。研究结果表明，本文所提出框架表现出优异且稳定的预测性能，在碳市场价格预测研究中具有良好的适用性和有效性。     

Impact of Reseller's Forecasting Accuracy on Channel Member Performance

This article studies a three‐layer supply chain where a manufacturer sells a product through a reseller who then relies on its own salesperson to sell to the end market. The reseller has superior capability in demand forecasting relative to the manufacturer. We explore the main trade‐offs between the risk‐reduction effect and the information–asymmetry–aggravation effect of the improved forecasting accuracy. We show that under the optimal wholesale price contract, both the manufacturer and the reseller are always better off as the reseller's forecasting accuracy improves. Nevertheless, under the menu of two‐part tariffs, the manufacturer prefers the reseller to be either uninformed or perfectly informed about the market condition. We further find that the improved forecasting accuracy is beneficial for the reseller if its current forecasting system is either very poor or very good.     

Impact of temporal aggregation on stock control performance of intermittent demand estimators: Empirical analysis

Intermittent demand is characterized by occasional demand arrivals interspersed by time intervals during which no demand occurs. These demand patterns pose considerable difficulties in terms of forecasting and stock control due to their compound nature, which implies variability both in terms of demand arrivals and demand sizes. An intuitively appealing strategy to deal with such patterns from a forecasting and stock control perspective is to aggregate demand in lower-frequency ‘time buckets’, thereby reducing the presence of zero observations. In this paper, we investigate the impact of forecasting aggregation on the stock control performance of intermittent demand patterns. The benefit of the forecasting aggregation approach is empirically assessed by means of analysis on a large demand dataset from the Royal Air Force (UK). The results show that the aggregation forecasting approach results in higher achieved service levels as compared to the classical forecasting approach. Moreover, when the combined service-cost performance is considered, the results also show that the former approach is more efficient than the latter, especially for high target service levels.     

基于PSO-PLS的组合预测方法在GDP预测中的应用

GDP预测是经济预测中一个非常重要的问题,随着经济的发展,对其预测精度的要求也越来越高.在考虑样本权重的基础上,提出一种微粒群算法与部分最小二乘回归方法相结合的组合预测方法,即采用微粒群方法对样本最优权重进行求解,在所得样本权重系数的基础上,用部分最小二乘回归方法确定组合预测的权重系数.将该方法用于中国GDP预测取得了较好的结果,与其他几种传统方法相比,预测精度有一定程度的提高,说明算法的有效性和可行性.     

Accuracy in the long run—Where are we now

Long-term forecasting must be viewed as informative speculation about the future. It should be credited with relatively little scientific authenticity. At the present stage of theory and data, no magic methods can be expected to overcome the problem of satisfactory forecasting. Improvement, however, is certainly not just a question of putting more and more variables and more and more numbers into a computer. Methods which indicate how to cushion against uncertainty and methods which bring a greater awareness of options for the future are essential if we are to gain greater control over events. The hazy images of the long-term future which are generated by scenario and other forms of analysis form the guidelines within which short and especially medium-term choices must be made. But in the end we must recognise the inherent limitations of forecasting and forecasting methods and think of forecasting not so much as a method of prediction but as a contribution to tackling the future in a more integrated sense.     

汇率的非线性组合预测方法研究

近年来的经济统计研究表明,组合预测比单项预测具有更高的预测精度,但线性组合预测方法在汇率的组合建模与预测方面存在着较大的局限性。本文提出了一种基于模糊神经网络的汇率非线性组合建模与预测新方法,并给出了相应的混合学习算法。对于英镑、法朗、瑞士法朗、日本元对美元等汇率时间序列的组合建模与预测结果表明,该方法具有很强的学习与泛化能力,在处理外汇市场这种具有一定程度不确定性的非线性系统的组合建模与预测方面有很好的应用价值。