A fast and scalable ensemble of global models with long memory and data partitioning for the M5 forecasting competition

This work presents key insights on the model development strategies used in our cross-learning-based retail demand forecast framework. The proposed framework outperforms state-of-the-art univariate models in the time series forecasting literature. It has achieved 17th position in the accuracy track of the M5 forecasting competition, which is among the top 1% of solutions.     

Improving forecasting performance using covariate-dependent copula models

Copulas provide an attractive approach to the construction of multivariate distributions with flexible marginal distributions and different forms of dependences. Of particular importance in many areas is the possibility of forecasting the tail-dependences explicitly. Most of the available approaches are only able to estimate tail-dependences and correlations via nuisance parameters, and cannot be used for either interpretation or forecasting. We propose a general Bayesian approach for modeling and forecasting tail-dependences and correlations as explicit functions of covariates, with the aim of improving the copula forecasting performance. The proposed covariate-dependent copula model also allows for Bayesian variable selection from among the covariates of the marginal models, as well as the copula density. The copulas that we study include the Joe-Clayton copula, the Clayton copula, the Gumbel copula and the Student’s $t$-copula. Posterior inference is carried out using an efficient MCMC simulation method. Our approach is applied to both simulated data and the S&P 100 and S&P 600 stock indices. The forecasting performance of the proposed approach is compared with those of other modeling strategies based on log predictive scores. A value-at-risk evaluation is also performed for the model comparisons.     

Data-based mechanistic modelling and forecasting globally averaged surface temperature

The main objective of this paper it to model the dynamic relationship between global averaged measures of Total Radiative Forcing (RTF) and surface temperature, measured by the Global Temperature Anomaly (GTA), and then use this model to forecast the GTA. The analysis utilizes the Data-Based Mechanistic (DBM) approach to the modelling and forecasting where, in this application, the unobserved component model includes a novel hybrid Box-Jenkins stochastic model in which the relationship between RTF and GTA is based on a continuous time transfer function (differential equation) model. This model then provides the basis for short term, inter-annual to decadal, forecasting of the GTA, using a transfer function form of the Kalman Filter, which produces a good prediction of the ‘pause’ or ‘levelling’ in the temperature rise over the period 2000 to 2011. This derives in part from the effects of a quasi-periodic component that is modelled and forecast by a Dynamic Harmonic Regression (DHR) relationship and is shown to be correlated with the Atlantic Multidecadal Oscillation (AMO) index.     

Big data driven cycle time parallel prediction for production planning in wafer manufacturing

Cycle time forecasting (CTF) is one of the most crucial issues for production planning to keep high delivery reliability in semiconductor wafer fabrication systems (SWFS). This paper proposes a novel data-intensive cycle time (CT) prediction system with parallel computing to rapidly forecast the CT of wafer lots with large datasets. First, a density peak based radial basis function network (DP-RBFN) is designed to forecast the CT with the diverse and agglomerative CT data. Second, the network learning method based on a clustering technique is proposed to determine the density peak. Third, a parallel computing approach for network training is proposed in order to speed up the training process with large scaled CT data. Finally, an experiment with respect to SWFS is presented, which demonstrates that the proposed CTF system can not only speed up the training process of the model but also outperform the radial basis function network, the back-propagation-network and multivariate regression methodology based CTF methods in terms of the mean absolute deviation and standard deviation.     

Forecasting Bordeaux wine prices using state-space methods

Generic Bordeaux red wine (basic claret) can be regarded as being similar to an agricultural commodity. Production volumes are substantial, they are traded at high frequency and the quality of the product is relatively homogeneous. Unlike other commodities and the top-end wines (which represent only 3% of the traded volume), there is no futures market for generic Bordeaux wine. Reliable forecasts of prices can to large extent replace this information deficiency and improve the functioning of the market. We use state-space methods with monthly data to obtain a univariate forecasting model for the average price. The estimates highlight the stochastic trend and the seasonality present in the evolution of the price over the period 1999 to 2016. The model predicts the path of wine prices out of sample reasonably well, suggesting that this approach is useful for making reasonably accurate forecasts of future price movements.     

Volatility forecasting: long memory,regime switching and heteroscedasticity

In this article, we account for the first time for long memory, regime switching and the conditional time-varying volatility of volatility (heteroscedasticity) to model and forecast market volatility using the heterogeneous autoregressive model of realized volatility (HAR-RV) and its extensions. We present several interesting and notable findings. First, existing models exhibit significant nonlinearity and clustering, which provide empirical evidence on the benefit of introducing regime switching and heteroscedasticity. Second, out-of-sample results indicate that combining regime switching and heteroscedasticity can substantially improve predictive power from a statistical viewpoint. More specifically, our proposed models generally exhibit higher forecasting accuracy. Third, these results are widely consistent across a variety of robustness tests such as different forecasting windows, forecasting models, realized measures, and stock markets. Consequently, this study sheds new light on forecasting future volatility.     

Using machine learning and big data for efficient forecasting of hotel booking cancellations

Cancellations are a key aspect of hotel revenue management because of their impact on room reservation systems. In fact, very little is known about the reasons that lead customers to cancel, or how it can be avoided. The aim of this paper is to propose a means of enabling the forecasting of hotel booking cancellations using only 13 independent variables, a reduced number in comparison with related research in the area, which in addition coincide with those that are most often requested by customers when they place a reservation. For this matter, machine-learning techniques, among other artificial neural networks optimised with genetic algorithms were applied achieving a cancellation rate of up to 98%. The proposed methodology allows us not only to know about cancellation rates, but also to identify which customer is likely to cancel. This approach would mean organisations could strengthen their action protocols regarding tourist arrivals.     

Paired comparison models with age effects modeled as piecewise quadratic splines

We propose new models for analyzing pairwise comparison data, such as that relating to sports. We focus on changes in players’ strengths and the prediction of future results. Our models are based on the Thurstone-Mosteller and Bradley–Terry models, and make use of the time variation in the parameters. Furthermore, we apply our models to data from the Japanese traditional sport sumo, and analyze this data. The proposed models perform better than the standard Thurstone-Mosteller and Bradley–Terry models according to both the Akaike information criterion and the Brier score. We compare the proposed models in detail by focusing on individual sumo wrestlers.     

Global energy forecasting competition 2017: Hierarchical probabilistic load forecasting

The Global Energy Forecasting Competition 2017 (GEFCom2017) attracted more than 300 students and professionals from over 30 countries for solving hierarchical probabilistic load forecasting problems. Of the series of global energy forecasting competitions that have been held, GEFCom2017 is the most challenging one to date: the first one to have a qualifying match, the first one to use hierarchical data with more than two levels, the first one to allow the usage of external data sources, the first one to ask for real-time ex-ante forecasts, and the longest one. This paper introduces the qualifying and final matches of GEFCom2017, summarizes the top-ranked methods, publishes the data used in the competition, and presents several reflections on the competition series and a vision for future energy forecasting competitions.     

How justified is abandoning money in the conduct of monetary policy in Australia on the grounds of instability in the money‐demand function?

The role of money in the design and conduct of monetary policy has reemerged as an important issue in both advanced and developing economies, especially since the 2007 global financial crisis. A growing body of recent literature suggests that the causal relationship between money supply growth and inflation remains intact across countries and over time and that this relation is not conditional on the stability of the money‐demand function or whether money is endogenous or exogenous. Moreover, critical for a rule‐based monetary policy is the presence of a long‐run stable money‐demand function, rather than a short‐run money‐demand model that may exhibit instability for many reasons, including problems with estimating a money‐demand model with high‐frequency data. Provided that a stable money‐demand function exists, it could be useful to establish long‐run equilibrium relations among money, output, prices, and exchange rates, as the classical monetary theory suggests. Within this analytical framework, this paper addresses the question of whether money has any role in the conduct of monetary policy in Australia. The conventional wisdom is that the money‐demand function in Australia has been unstable since the mid‐1980s due to financial deregulation and reforms; this led to a change in the strategy of monetary policy for price stability in the form of inflation targeting that ignores money insofar as inflation and its control are concerned. This paper reports empirical findings for Australia, obtained from a longer quarterly data series over the period 1960Q1–2015Q1, which suggest that instability in the narrow‐money‐demand function in Australia was primarily due to the exclusion of variables which have become important in the deregulated environment since the 1980s. These findings are confirmed by an expanded form of the narrow‐money‐demand function that was found stable over the past two decades, although it experienced multiple structural breaks over the study period. The paper draws the conclusion that abandoning the monetary aggregate as an instrument of monetary policy in Australia, under a rule‐based monetary policy such as inflation targeting, cannot be justified by instability in the money‐demand function or even by lack of a causal link between money supply growth and inflation.