International Journal of Forecasting最新文献

The National Hockey League (NHL) Entry Draft has been an active area of research in hockey analytics over the past decade. Prior research has explored predictive modelling for draft results using player information and statistics as well as ranking data from draft experts. In this paper, we develop a new modelling framework for this problem using a Bayesian rank-ordered logit model based on draft ranking data from industry experts between 2019 and 2022. This model builds upon previous approaches by incorporating team tendencies, addressing within-ranking dependence between players, and solving various other challenges of working with rank-ordered outcomes, such as incorporating both unranked players and rankings that only consider a subset of the available pool of players (e.g., North American skaters, European goalies, etc.).

This paper uses deep learning techniques to conduct a systematic large-scale analysis of order book-driven predictability in high-frequency returns. First, we introduce a new and robust representation of the order book, the volume representation. Next, we conduct an extensive empirical experiment to address various questions regarding predictability. We investigate if and how far ahead there is predictability, the importance of a robust data representation, the advantages of multi-horizon modeling, and the presence of universal trading patterns. We use model confidence sets, which provide a formalized statistical inference framework well suited to answer these questions. Our findings show that at high frequencies, predictability in mid-price returns is not just present but ubiquitous. The performance of the deep learning models is strongly dependent on the choice of order book representation, and in this respect, the volume representation appears to have multiple practical advantages.

This paper presents a new method for combining (or aggregating or ensembling) multivariate probabilistic forecasts, considering dependencies between quantiles and marginals through a smoothing procedure that allows for online learning. We discuss two smoothing methods: dimensionality reduction using Basis matrices and penalized smoothing. The new online learning algorithm generalizes the standard CRPS learning framework into multivariate dimensions. It is based on Bernstein Online Aggregation (BOA) and yields optimal asymptotic learning properties. The procedure uses horizontal aggregation, i.e., aggregation across quantiles. We provide an in-depth discussion on possible extensions of the algorithm and several nested cases related to the existing literature on online forecast combination. We apply the proposed methodology to forecasting day-ahead electricity prices, which are 24-dimensional distributional forecasts. The proposed method yields significant improvements over uniform combination in terms of continuous ranked probability score (CRPS). We discuss the temporal evolution of the weights and hyperparameters and present the results of reduced versions of the preferred model. A fast C＋＋implementation of the proposed algorithm is provided in the open-source R-Package profoc on CRAN.

In the hospitality industry, lead time data are a form of compositional data that are crucial for business planning, resource allocation, and staffing. Hospitality businesses accrue fees daily, but recognition of these fees is often deferred. This paper presents a novel class of Bayesian time series models, the Bayesian Dirichlet auto-regressive moving average (B-DARMA) model, designed specifically for compositional time series. The model is motivated by the analysis of five years of daily fees data from Airbnb, with the aim of forecasting the proportion of future fees that will be recognized in 12 consecutive monthly intervals. Each day’s compositional data are modeled as Dirichlet distributed, given the mean and a scale parameter. The mean is modeled using a vector auto-regressive moving average process, which depends on previous compositional data, previous compositional parameters, and daily covariates. The B-DARMA model provides a robust solution for analyzing large compositional vectors and time series of varying lengths. It offers efficiency gains through the choice of priors, yields interpretable parameters for inference, and produces reasonable forecasts. The paper also explores the use of normal and horseshoe priors for the vector auto-regressive and vector moving average coefficients, and for regression coefficients. The efficacy of the B-DARMA model is demonstrated through simulation studies and an analysis of Airbnb data.

We forecast CPI inflation indicators in the United Kingdom using a large set of monthly disaggregated CPI item series covering a sample period of twenty years, and employing a range of forecasting tools to deal with the high dimension of the set of predictors. Although an autoregressive model proofs hard to outperform overall, Ridge regression combined with CPI item series performs strongly in forecasting headline inflation. A range of shrinkage methods yields significant improvement over sub-periods where inflation was rising, falling or in the tails of its distribution. Once CPI item series are exploited, we find little additional forecast gain from including macroeconomic predictors. The forecast performance of non-parametric machine learning methods is relatively weak. Using Shapley values to decompose forecast signals exploited by a Random Forest, we show that the ability of non-parametric tools to flexibly switch between signals from groups of indicators may come at the cost of high variance and, as such, hurt forecast performance.

This paper presents a dynamic network autoregressive conditional heteroscedasticity (ARCH) model suitable for high-dimensional cases where multivariate ARCH models are typically no longer applicable. We adopt the theoretical foundations from spatiotemporal statistics and transfer the dynamic ARCH model processes to networks. The model integrates temporally lagged volatility and information from adjacent nodes, which may instantaneously spill across the entire network. The model is used to forecast volatility in the US stock market, and the edges are determined based on various distance and correlation measures between the time series. The performance of alternative network definitions is compared with independent univariate log-ARCH models in terms of out-of-sample prediction accuracy. The results indicate that more accurate forecasts are obtained with network-based models and that accuracy can be improved by combining the forecasts of different network definitions. We emphasise the significance for practitioners to integrate network structure information when developing volatility forecasts.

Multi-step prediction is a key challenge in univariate forecasting. However, forecasting accuracy decreases as predictions are made further into the future. This is caused by the decreasing predictability and the error propagation along the horizon. In this paper, we propose a novel method called Forecasted Trajectory Neighbors (FTN) for multi-step forecasting with univariate time series. FTN is a meta-learning strategy that can be integrated with any state-of-the-art multi-step forecasting approach. It works by using training observations to correct the errors made during multiple predictions. This is accomplished by retrieving the nearest neighbors of the multi-step forecasts and averaging these for prediction. The motivation is to introduce, in a lightweight manner, a conditional dependent constraint across the forecasting horizons. Such a constraint, not always taken into account by most strategies, can be considered as a sort of regularization element. We carried out extensive experiments using 7795 time series from different application domains. We found that our method improves the performance of several state-of-the-art multi-step forecasting methods. An implementation of the proposed method is publicly available online, and the experiments are reproducible.

We apply fixed-$b$ and fixed-$m$ asymptotics to tests of equal predictive accuracy and of encompassing for survey density forecasts. We verify in an original Monte Carlo design that fixed-smoothing asymptotics delivers correctly sized tests in this framework, even when only a small number of out of sample observations is available. We use the proposed density forecast comparison tests with fixed-smoothing asymptotics to assess the predictive ability of density forecasts from the European Central Bank’s Survey of Professional Forecasters (ECB SPF). We find an improvement in the relative predictive ability of the ECB SPF since 2010, suggesting a change in the forecasting practice after the financial crisis.