Time series forecasting with neural network ensembles: an application for exchange rate prediction

This paper investigates the use of neural network combining methods to improve time series forecasting performance of the traditional single keep-the-best (KTB) model. The ensemble methods are applied to the difficult problem of exchange rate forecasting. Two general approaches to combining neural networks are proposed and examined in predicting the exchange rate between the British pound and US dollar. Specifically, we propose to use systematic and serial partitioning methods to build neural network ensembles for time series forecasting. It is found that the basic ensemble approach created with non-varying network architectures trained using different initial random weights is not effective in improving the accuracy of prediction while ensemble models consisting of different neural network structures can consistently outperform predictions of the single ‘best’ network. Results also show that neural ensembles based on different partitions of the data are more effective than those developed with the full training data in out-of-sample forecasting. Moreover, reducing correlation among forecasts made by the ensemble members by utilizing data partitioning techniques is the key to success for the neural ensemble models. Although our ensemble methods show considerable advantages over the traditional KTB approach, they do not have significant improvement compared to the widely used random walk model in exchange rate forecasting.     

Short-run electricity load forecasting with combinations of stationary wavelet transforms

Short-term forecasting of electricity load is an essential issue for the management of power systems and for energy trading. Specific modeling approaches are needed given the strong seasonality and volatility in load data. In this paper, we investigate the benefit of combining stationary wavelet transforms to produce one day-ahead forecasts of half-hourly electric load in France. First, we assess the advantage of decomposing the aggregate load into several subseries with a wavelet transform. Each component is predicted separately and aggregated to get the final forecast. One innovation of this paper is to propose several approaches to deal with the boundary problem which is particularly detrimental in electricity load forecasting. Second, we examine the benefit of combining forecasts over individual models. An extensive out-of-sample evaluation shows that a careful treatment of the border effect is required in the multiresolution analysis. Combinations including the wavelet predictions provide the most accurate forecasts. This result is valid with several assumptions about the forecast error in temperature and for different types of hours (peak, normal, off-peak), different days of the week and various forecasting periods.     

Volatility forecasting via SVR–GARCH with mixture of Gaussian kernels

The support vector regression (SVR) is a supervised machine learning technique that has been successfully employed to forecast financial volatility. As the SVR is a kernel-based technique, the choice of the kernel has a great impact on its forecasting accuracy. Empirical results show that SVRs with hybrid kernels tend to beat single-kernel models in terms of forecasting accuracy. Nevertheless, no application of hybrid kernel SVR to financial volatility forecasting has been performed in previous researches. Given that the empirical evidence shows that the stock market oscillates between several possible regimes, in which the overall distribution of returns it is a mixture of normals, we attempt to find the optimal number of mixture of Gaussian kernels that improve the one-period-ahead volatility forecasting of SVR based on GARCH(1,1). The forecast performance of a mixture of one, two, three and four Gaussian kernels are evaluated on the daily returns of Nikkei and Ibovespa indexes and compared with SVR–GARCH with Morlet wavelet kernel, standard GARCH, Glosten–Jagannathan–Runkle (GJR) and nonlinear EGARCH models with normal, student-t, skew-student-t and generalized error distribution (GED) innovations by using mean absolute error (MAE), root mean squared error (RMSE) and robust Diebold–Mariano test. The results of the out-of-sample forecasts suggest that the SVR–GARCH with a mixture of Gaussian kernels can improve the volatility forecasts and capture the regime-switching behavior.     

Forecasting foreign exchange rates with adaptive neural networks using radial-basis functions and Particle Swarm Optimization

The motivation for this paper is to introduce a hybrid neural network architecture of Particle Swarm Optimization and Adaptive Radial Basis Function (ARBF–PSO), a time varying leverage trading strategy based on Glosten, Jagannathan and Runkle (GJR) volatility forecasts and a neural network fitness function for financial forecasting purposes. This is done by benchmarking the ARBF–PSO results with those of three different neural networks architectures, a Nearest Neighbors algorithm (k-NN), an autoregressive moving average model (ARMA), a moving average convergence/divergence model (MACD) plus a na?¨ve strategy. More specifically, the trading and statistical performance of all models is investigated in a forecast simulation of the EUR/USD, EUR/GBP and EUR/JPY ECB exchange rate fixing time series over the period January 1999–March 2011 using the last 2 years for out-of-sample testing.     

Forecasting tender price index under incomplete information

Tender price index (TPI) is essential for estimating the likely tender price of a given project. Due to incomplete information on future market conditions, it is difficult to accurately forecast the TPI. Most traditional statistical forecasting models require a certain number of historical data, which may not be completely available in many practical situations. In order to overcome this problem, the grey model is proposed for forecasting TPIs because it only requires a small number of input data. For this study, the data source was based on the TPIs produced by the Government's Architectural Services Department. On the basis of four input data, the grey model forecasted TPIs from 1981Q1 to 2011Q4. The mean absolute percentage errors of forecast TPIs in one quarter and two quarters ahead were 3.62 and 7.04%, respectively. In order to assess the accuracy and reliability of the grey model further, the same research method was used to forecast other three TPIs in Hong Kong. The forecasting results of all four TPIs were found to be very good. It was thus concluded that the grey model could be able to produce accurate TPI forecasts for a one-quarter to two-quarter forecast horizon.     

A short and mean-term automatic forecasting system––application to textile logistics

In order to reduce their stocks and to limit stock out, textile companies require specific and accurate sale forecasting systems. More especially, textile distribution involves different forecast lead times: mean-term (one year) and short-term (one week in average). This paper presents two new complementary forecasting models, appropriate to textile market requirements. The first model (AHFCCX) allows to automatically obtain mean-term forecasting by using fuzzy techniques to quantify influence of explanatory variables. The second one (SAMANFIS), based on a neuro-fuzzy method, performs short-term forecasting by readjusting mean-term model forecasts from load real sales. To evaluate forecasts accuracy, our models and classical ones are compared to 322 real items sales series of an important ready to wear distributor.     

A Theoretically Consistent Version of the Nelson and Siegel Class of Yield Curve Models

A popular class of yield curve models is based on the Nelson and Siegel approach of ‘fitting’ yield curve data with simple functions of maturity. However, such models cannot be consistent across time. This article addresses that deficiency by deriving an intertemporally consistent and arbitrage‐free version of the Nelson and Siegel model. Adding this theoretical consistency expands the potential applications of the Nelson and Siegel approach to exercises involving a time‐series context, such as forecasting the yield curve and pricing interest rate derivatives. As a practical example, the intertemporal consistency of the model is exploited to derive a theoretical framework for forecasting the yield curve. The empirical application of that framework to United States data results in out‐of‐sample forecasts that outperform the random walk over the sample period of almost 50 years, for forecast horizons ranging from six months to three years.     

Electricity price forecasting accounting for renewable energies: optimal combined forecasts

Electricity price forecasting is an interesting problem for all the agents involved in electricity market operation. For instance, every profit maximisation strategy is based on the computation of accurate one-day-ahead forecasts, which is why electricity price forecasting has been a growing field of research in recent years. In addition, the increasing concern about environmental issues has led to a high penetration of renewable energies, particularly wind. In some European countries such as Spain, Germany and Denmark, renewable energy is having a deep impact on the local power markets. In this paper, we propose an optimal model from the perspective of forecasting accuracy, and it consists of a combination of several univariate and multivariate time series methods that account for the amount of energy produced with clean energies, particularly wind and hydro, which are the most relevant renewable energy sources in the Iberian Market. This market is used to illustrate the proposed methodology, as it is one of those markets in which wind power production is more relevant in terms of its percentage of the total demand, but of course our method can be applied to any other liberalised power market. As far as our contribution is concerned, first, the methodology proposed by García-Martos et al (2007 and 2012) is generalised twofold: we allow the incorporation of wind power production and hydro reservoirs, and we do not impose the restriction of using the same model for 24?h. A computational experiment and a Design of Experiments (DOE) are performed for this purpose. Then, for those hours in which there are two or more models without statistically significant differences in terms of their forecasting accuracy, a combination of forecasts is proposed by weighting the best models (according to the DOE) and minimising the Mean Absolute Percentage Error (MAPE). The MAPE is the most popular accuracy metric for comparing electricity price forecasting models. We construct the combination of forecasts by solving several nonlinear optimisation problems that allow computation of the optimal weights for building the combination of forecasts. The results are obtained by a large computational experiment that entails calculating out-of-sample forecasts for every hour in every day in the period from January 2007 to December 2009. In addition, to reinforce the value of our methodology, we compare our results with those that appear in recent published works in the field. This comparison shows the superiority of our methodology in terms of forecasting accuracy.     

Unsupervised extreme learning machine and support vector regression hybrid model for predicting energy commodity futures index

Support vector regression (SVR) has been successfully applied in various domains, including predicting the prices of different financial instruments like stocks, futures, options, and indices. Because of the wide variation in financial time-series data, instead of using only a single standard prediction technique like SVR, we propose a hybrid model called USELM-SVR. It is a combination of unsupervised extreme learning machine (US-ELM)-based clustering and SVR forecasting. We assessed the feasibility and effectiveness of this hybrid model using a case study, predicting the one-, two-, and three-day ahead closing values of the energy commodity futures index traded on the Multi Commodity Exchange in India. Our experimental results show that the USELM-SVR is viable and effective, and produces better forecasts than our benchmark models (standard SVR, a hybrid of SVR with self-organizing map (SOM) clustering, and a hybrid of SVR with k-means clustering). Moreover, the proposed USELM-SVR architecture is useful as an alternative model for prediction tasks when we require more accurate predictions.     

Mortality density forecasts: An analysis of six stochastic mortality models

This paper develops a framework for developing forecasts of future mortality rates. We discuss the suitability of six stochastic mortality models for forecasting future mortality and estimating the density of mortality rates at different ages. In particular, the models are assessed individually with reference to the following qualitative criteria that focus on the plausibility of their forecasts: biological reasonableness; the plausibility of predicted levels of uncertainty in forecasts at different ages; and the robustness of the forecasts relative to the sample period used to fit the model. An important, though unsurprising, conclusion is that a good fit to historical data does not guarantee sensible forecasts. We also discuss the issue of model risk, common to many modelling situations in demography and elsewhere. We find that even for those models satisfying our qualitative criteria, there are significant differences among central forecasts of mortality rates at different ages and among the distributions surrounding those central forecasts.     

Weibull分布在新产品市场渗透研究中的应用拓展

在应用Weibull模型研究新产品市场渗透时,"永不采用人群"、消费者个体之间的差异、消费者群体之间的差异是研究新产品采用时需要考虑的三种因素。本文基于这三种因素分别建立了三个拓展的Weibull模型,并利用面板数据进行了实证研究,发现三种拓展之后的模型在数据拟合和数据预测方面均有显著的提高.然后,本文将三种因素整合至一个模型之中形成了一种新的综合Weibull模型,实证分析结果显示新的模型有很好的新产品市场渗透数据拟合和预测能力.     

The quality of bank forecasts: The dollar-pound exchange rate, 1990–1993

A general framework for examining the quality of currency forecasts is described. This framework incorporates existing accuracy measures and modifies them to give components of accuracy that are appropriate for statistically derived and judgementally based forecasts. The framework is described and applied to one week ahead US$/UK£ forecasts from three major banks over a three year period between 1990 and 1993. The results suggest that, while overall forecast performance was poor, some aspects of the predictions could still be useful in a practical setting.     

Estimating VAR models for the term structure of interest rates

In this paper we follow the work of Evans and Marshall and propose new approaches for modelling the joint development of macro variables and the returns of government bond yields of several maturities. The models are estimated and compared with other forecasting schemes previously proposed in the literature, especially those relying on univariate, VAR and error correction methods. The models are then used to judge the hypothesis that the information content of macro variables and the term structure of interest rates as a whole help improving forecasting performance. Our main conclusion is quite simple: if one is interested in computing short-term forecasts, then there is no significant improvement in incorporating information other than the one already present in past observations of the yield at hand; however, if one worries about long-term forecasts (which is frequently the case with pension insurance companies), then the information content of macro variables and the term structure can improve forecasting performance.     

Using shared sell-through data to forecast wholesaler demand in multi-echelon supply chains

Operational forecasting in supply chain management supports a variety of short-term planning decisions, such as production scheduling and inventory management. In this respect, improving short-term forecast accuracy is a way to build a more agile supply chain for manufacturing companies. Demand forecasting often relies on well-established univariate forecasting methods to extrapolate historical demand. Collaboration across the supply chain, including information sharing, is suggested in the literature to improve upon the forecast accuracy of such traditional methods. In this paper, we review empirical studies considering the use of downstream information in demand forecasting and investigate different modeling approaches and forecasting methods to incorporate such data. Where empirical findings on information sharing mainly focus on point-of-sale data in two-level supply chains, this research empirically investigates the added value of using sell-through data originating from intermediaries, next to historical demand figures, in a multi-echelon supply chain. In a case study concerning a US drug manufacturer, we evaluate different methods to incorporate this data and consider both time series methods and machine learning techniques to produce multi-step ahead weekly forecasts. The results show that the manufacturer can effectively improve its short-term forecast accuracy by integrating sell-through data into the forecasting process and provide useful insights as to the different modeling approaches used. The conclusion holds for all forecast horizons considered, though it is most pronounced for one-step ahead forecasts. Therefore, our research provides a clear incentive for manufacturers to assess the forecast accuracy that can be achieved by using sell-through data.     

不同矩属性波动模型对中国股市波动率的预测精度分析

金融时间序列的波动性建模经历了从一阶矩到二阶矩直到高阶矩(包含三阶矩和四阶矩)的过程,而对于高阶矩波动模型是否有助于对未来市场的波动率预测这一问题,国内外学术界尚无文献讨论。以上证综指长达7年的每5分钟高频数据样本为例,通过构建具有不同矩属性的波动模型,计算了中国股票市场波动率的预测值,并利用具有bootstrap特性的SPA检验法,实证检验了不同矩属性波动模型的波动率预测精度差异。实证结果显示:就中国股市而言,四阶矩波动模型能够取得比二阶矩波动模型更优的波动率预测精度,而三阶矩波动模型并未表现出比二阶矩波动模型更强的预测能力;在高阶矩波动模型中包含杠杆效应项并不能提高模型的预测精度。最后提出了在金融风险管理、衍生产品定价等领域引入四阶矩波动模型的研究思路。     

Potential inventory cost reductions using advanced time series forecasting techniques

This paper compares demand forecasts computed using the time series forecasting techniques of vector autoregression (VAR) and Bayesian VAR (BVAR) with forecasts computed using exponential smoothing and seasonal decomposition. These forecasts for three demand data series were used to determine three inventory management policies for each time series. The inventory costs associated with each of these policies were used as a further basis for comparison of the forecasting techniques. The results show that the BVAR technique, which uses mixed estimation, is particularly useful in reducing inventory costs in cases where the limited historical data offer little useful information for forecasting. The BVAR technique was effective in improving forecast accuracy and reducing inventory costs in two of the three cases tested. In the third case, unrestricted VAR and exponential smoothing produced the lowest experimental forecast errors and computed inventory costs. Furthermore, this research illustrates that improvements in demand forecasting can provide better cost reductions than relying on stochastic inventory models to provide cost reductions.     

Forecasting white goods in a recesion

The forecasting models discussed in this paper deal with thesales prospects of three categories of an unspecified whitegood. The forecasts pertain to UK deliveries both of a particularfirm and of all UK suppliers. The forecasting approach essentiallyresorts to multiple-regression models that employ leading indicatorswith liberal resort to polynomial distributed lags. Considerablestructural differences in these models as between firm and industryreflect distinctive features of the firm's product line andpolicies. The models are first analysed in terms of their predictiveperformance, bothin-sample and ex-sample. Features of thesemodels and of their predictions are then considered in termsof what they indicate about the firm's policy options.     

基于贝叶斯向量自回归的区域经济预测模型:以青海为例

由于缺乏足够的观测数据等原因,常规的区域经济预测模型在我国难以获得预期的预测效果,而贝叶斯向量自回归(BVAR)模型将变量的统计性质作为参数的先验分布引入到传统的VAR模型中,能够克服自由度过少的问题,以青海为例,本文建立了一个BVAR模型,并引入了全国GDP和中央政府转移支付作为外生变量以描述国民经济与区域经济的联系...     

A comparison of the Lee-Carter model and AR-ARCH model for forecasting mortality rates

With the decline in the mortality level of populations, national social security systems and insurance companies of most developed countries are reconsidering their mortality tables taking into account the longevity risk. The Lee and Carter model is the first discrete-time stochastic model to consider the increased life expectancy trends in mortality rates and is still broadly used today. In this paper, we propose an alternative to the Lee-Carter model: an AR(1)-ARCH(1) model. More specifically, we compare the performance of these two models with respect to forecasting age-specific mortality in Italy. We fit the two models, with Gaussian and t-student innovations, for the matrix of Italian death rates from 1960 to 2003. We compare the forecast ability of the two approaches in out-of-sample analysis for the period 2004-2006 and find that the AR(1)-ARCH(1) model with t-student innovations provides the best fit among the models studied in this paper.     

Forecasting mortality rate improvements with a high-dimensional VAR

Forecasting mortality rates is a problem which involves the analysis of high-dimensional time series. Most of usual mortality models propose to decompose the mortality rates into several latent factors to reduce this complexity. These approaches, in particular those using cohort factors, have a good fit, but they are less reliable for forecasting purposes. One of the major challenges is to determine the spatial–temporal dependence structure between mortality rates given a relatively moderate sample size. This paper proposes a large vector autoregressive (VAR) model fitted on the differences in the log-mortality rates, ensuring the existence of long-run relationships between mortality rate improvements. Our contribution is threefold. First, sparsity, when fitting the model, is ensured by using high-dimensional variable selection techniques without imposing arbitrary constraints on the dependence structure. The main interest is that the structure of the model is directly driven by the data, in contrast to the main factor-based mortality forecasting models. Hence, this approach is more versatile and would provide good forecasting performance for any considered population. Additionally, our estimation allows a one-step procedure, as we do not need to estimate hyper-parameters. The variance–covariance matrix of residuals is then estimated through a parametric form. Secondly, our approach can be used to detect nonintuitive age dependence in the data, beyond the cohort and the period effects which are implicitly captured by our model. Third, our approach can be extended to model the several populations in long run perspectives, without raising issue in the estimation process. Finally, in an out-of-sample forecasting study for mortality rates, we obtain rather good performances and more relevant forecasts compared to classical mortality models using the French, US and UK data. We also show that our results enlighten the so-called cohort and period effects for these populations.