Evaluating quantile forecasts in the M5 uncertainty competition

Probabilistic forecasts are necessary for robust decisions in the face of uncertainty. The M5 Uncertainty competition required participating teams to forecast nine quantiles for unit sales of various products at various aggregation levels and for different time horizons. This paper evaluates the forecasting performance of the quantile forecasts at different aggregation levels and at different quantile levels. We contrast this with some theoretical predictions, and discuss potential implications and promising future research directions for the practice of probabilistic forecasting.     

A strategic predictive distribution for tests of probabilistic calibration

Forecasts of probability distributions are needed to support decision making in many applications. The accuracy of predictive distributions should be evaluated by maximising sharpness subject to calibration. Sharpness relates to the concentration of the predictive distributions, while calibration concerns their statistical consistency with the data. This paper focuses on calibration testing. It is important that a calibration test cannot be gamed by forecasts that have been strategically designed to pass the test. The widely used tests of probabilistic calibration for predictive distributions are based on the probability integral transform. Drawing on previous results for quantile prediction, we show that strategic distributional forecasting is a concern for these tests. To address this, we provide a simple extension of one of the tests. We illustrate ideas using simulated data.     

A copula-based time series model for global horizontal irradiation

The increasing importance of solar power for electricity generation leads to increasing demand for probabilistic forecasting of local and aggregated photovoltaic (PV) yields. Based on publicly available irradiation data, this paper uses an indirect modeling approach for hourly medium to long-term local PV yields. We suggest a time series model for global horizontal irradiation that allows for multivariate probabilistic forecasts for arbitrary time horizons. It features several important stylized facts. Sharp time-dependent lower and upper bounds of global horizontal irradiations are estimated. The parameters of the beta distributed marginals of the transformed data are allowed to be time-dependent. A copula-based time series model is introduced for the hourly and daily dependence structure based on simple vine copulas with so-called tail dependence. Evaluation methods based on scoring rules are used to compare the model’s power for multivariate probabilistic forecasting with other models used in the literature showing that our model outperforms other models in many respects.     

The performance of the global bottom-up approach in the M5 accuracy competition: A robustness check

The M5 accuracy competition has presented a large-scale hierarchical forecasting problem in a realistic grocery retail setting in order to evaluate an extended range of forecasting methods, particularly those adopting machine learning. The top ranking solutions adopted a global bottom-up approach, by which is meant using global forecasting methods to generate bottom level forecasts in the hierarchy and then using a bottom-up strategy to obtain coherent forecasts for aggregate levels. However, whether the observed superior performance of the global bottom-up approach is robust over various test periods or only an accidental result, is an important question for retail forecasting researchers and practitioners. We conduct experiments to explore the robustness of the global bottom-up approach, and make comments on the efforts made by the top-ranking teams to improve the core approach. We find that the top-ranking global bottom-up approaches lack robustness across time periods in the M5 data. This inconsistent performance makes the M5 final rankings somewhat of a lottery. In future forecasting competitions, we suggest the use of multiple rolling test sets to evaluate the forecasting performance in order to reward robustly performing forecasting methods, a much needed characteristic in any application.     

M5 accuracy competition: Results,findings, and conclusions

In this study, we present the results of the M5 “Accuracy” competition, which was the first of two parallel challenges in the latest M competition with the aim of advancing the theory and practice of forecasting. The main objective in the M5 “Accuracy” competition was to accurately predict 42,840 time series representing the hierarchical unit sales for the largest retail company in the world by revenue, Walmart. The competition required the submission of 30,490 point forecasts for the lowest cross-sectional aggregation level of the data, which could then be summed up accordingly to estimate forecasts for the remaining upward levels. We provide details of the implementation of the M5 “Accuracy” challenge, as well as the results and best performing methods, and summarize the major findings and conclusions. Finally, we discuss the implications of these findings and suggest directions for future research.     

Forecasting with high frequency data: M4 competition and beyond

The M4 competition included 100,000 time series, with the frequencies ranging from yearly to hourly. The team rankings differ notably across frequencies for both point and probabilistic forecasting. I discuss the performances of these methods, with an emphasis on the hourly series of the M4 competition. I also discuss forecasting with high-frequency data in general.     

The M5 competition: Background,organization, and implementation

The M5 competition follows the previous four M competitions, whose purpose is to learn from empirical evidence how to improve forecasting performance and advance the theory and practice of forecasting. M5 focused on a retail sales forecasting application with the objective to produce the most accurate point forecasts for 42,840 time series that represent the hierarchical unit sales of the largest retail company in the world, Walmart, as well as to provide the most accurate estimates of the uncertainty of these forecasts. Hence, the competition consisted of two parallel challenges, namely the Accuracy and Uncertainty forecasting competitions. M5 extended the results of the previous M competitions by: (a) significantly expanding the number of participating methods, especially those in the category of machine learning; (b) evaluating the performance of the uncertainty distribution along with point forecast accuracy; (c) including exogenous/explanatory variables in addition to the time series data; (d) using grouped, correlated time series; and (e) focusing on series that display intermittency. This paper describes the background, organization, and implementations of the competition, and it presents the data used and their characteristics. Consequently, it serves as introductory material to the results of the two forecasting challenges to facilitate their understanding.     

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.     

A white-boxed ISSM approach to estimate uncertainty distributions of Walmart sales

We present our solution for the M5 Uncertainty competition. Our solution ranked sixth out of 909 submissions across all hierarchical levels and ranked first for prediction at the finest level of granularity (product-store sales, i.e. SKUs). The model combines a multi-stage state-space model and Monte Carlo simulations to generate the forecasting scenarios (trajectories). Observed sales are modelled with negative binomial distributions to represent discrete over-dispersed sales. Seasonal factors are handcrafted and modelled with linear coefficients that are calculated at the store-department level.     

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.     

The M5 uncertainty competition: Results,findings and conclusions

This paper describes the M5 “Uncertainty” competition, the second of two parallel challenges of the latest M competition, aiming to advance the theory and practice of forecasting. The particular objective of the M5 “Uncertainty” competition was to accurately forecast the uncertainty distributions of the realized values of 42,840 time series that represent the hierarchical unit sales of the largest retail company in the world by revenue, Walmart. To do so, the competition required the prediction of nine different quantiles (0.005, 0.025, 0.165, 0.250, 0.500, 0.750, 0.835, 0.975, and 0.995), that can sufficiently describe the complete distributions of future sales. The paper provides details on the implementation and execution of the M5 “Uncertainty” competition, presents its results and the top-performing methods, and summarizes its major findings and conclusions. Finally, it discusses the implications of its findings and suggests directions for future research.     

Kaggle forecasting competitions: An overlooked learning opportunity

We review the results of six forecasting competitions based on the online data science platform Kaggle, which have been largely overlooked by the forecasting community. In contrast to the M competitions, the competitions reviewed in this study feature daily and weekly time series with exogenous variables, business hierarchy information, or both. Furthermore, the Kaggle data sets all exhibit higher entropy than the M3 and M4 competitions, and they are intermittent.In this review, we confirm the conclusion of the M4 competition that ensemble models using cross-learning tend to outperform local time series models and that gradient boosted decision trees and neural networks are strong forecast methods. Moreover, we present insights regarding the use of external information and validation strategies, and discuss the impacts of data characteristics on the choice of statistics or machine learning methods. Based on these insights, we construct nine ex-ante hypotheses for the outcome of the M5 competition to allow empirical validation of our findings.     

DeepAR: Probabilistic forecasting with autoregressive recurrent networks

Probabilistic forecasting, i.e., estimating a time series’ future probability distribution given its past, is a key enabler for optimizing business processes. In retail businesses, for example, probabilistic demand forecasts are crucial for having the right inventory available at the right time and in the right place. This paper proposes DeepAR, a methodology for producing accurate probabilistic forecasts, based on training an autoregressive recurrent neural network model on a large number of related time series. We demonstrate how the application of deep learning techniques to forecasting can overcome many of the challenges that are faced by widely-used classical approaches to the problem. By means of extensive empirical evaluations on several real-world forecasting datasets, we show that our methodology produces more accurate forecasts than other state-of-the-art methods, while requiring minimal manual work.     

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.     

Applicability of the M5 to Forecasting at Walmart

The M5 Forecasting Competition, the fifth in the series of forecasting competitions organized by Professor Spyros Makridakis and the Makridakis Open Forecasting Center at the University of Nicosia, was an extremely successful event. This competition focused on both the accuracy and uncertainty of forecasts and leveraged actual historical sales data provided by Walmart. This has led to the M5 being a unique competition that closely parallels the difficulties and challenges associated with industrial applications of forecasting. Like its precursor the M4, many interesting ideas came from the results of the M5 competition which will continue to push forecasting in new directions.In this article we discuss four topics around the practitioners view of the application of the competition and its results to the actual problems we face. First, we examine the data provided and how it relates to common difficulties practitioners must overcome. Secondly, we review the relevance of the accuracy and uncertainty metrics associated with the competition. Third, we discuss the leading solutions and their implications to forecasting at a company like Walmart. We then close with thoughts about a future M6 competition and further enhancements that can be explored.     

Probabilistic wind forecasting up to three months ahead using ensemble predictions for geopotential height

We develop a method for forecasting the distribution of the daily surface wind speed at timescales from 15-days to 3-months in France. On such long-term timescales, ensemble predictions of the surface wind speed have poor performance, however, the wind speed distribution may be related to the large-scale circulation of the atmosphere, for which the ensemble forecasts have better skill. The information from the large-scale circulation, represented by the 500 hPa geopotential height, is summarized into a single index by first running a PCA and then a polynomial regression. We estimate, over 20 years of daily data, the conditional probability density of the wind speed at a specific location given the index. We then use the ECMWF seasonal forecast ensemble to predict the index for horizons from 15-days to 3-months. These predictions are plugged into the conditional density to obtain a distributional forecast of surface wind. These probabilistic forecasts remain sharper than the climatology up to 1-month forecast horizon. Using a statistical postprocessing method to recalibrate the ensemble leads to further improvement of our probabilistic forecast, which then remains calibrated and sharper than the climatology up to 3-months horizon, particularly in the north of France in winter and fall.     

Transfer learning for hierarchical forecasting: Reducing computational efforts of M5 winning methods

The winning machine learning methods of the M5 Accuracy competition demonstrated high levels of forecast accuracy compared to the top-performing benchmarks in the history of the M-competitions. Yet, large-scale adoption is hampered due to the significant computational requirements to model, tune, and train these state-of-the-art algorithms. To overcome this major issue, we discuss the potential of transfer learning (TL) to reduce the computational effort in hierarchical forecasting and provide a proof of concept that TL can be applied on M5 top-performing methods. We demonstrate our easy-to-use TL framework on the recursive store-level LightGBM models of the M5 winning method and attain similar levels of forecast accuracy with roughly 25% less training time. Our findings provide evidence for a novel application of TL to facilitate the practical applicability of the M5 winning methods in large-scale settings with hierarchically structured data.     

Probabilistic forecasting of industrial electricity load with regime switching behavior

This paper suggests a novel inhomogeneous Markov switching approach for the probabilistic forecasting of industrial companies’ electricity loads, for which the load switches at random times between production and standby regimes. The model that we propose describes the transitions between the regimes using a hidden Markov chain with time-varying transition probabilities that depend on calendar variables. We model the demand during the production regime using an autoregressive moving-average (ARMA) process with seasonal patterns, whereas we use a much simpler model for the standby regime in order to reduce the complexity. The maximum likelihood estimation of the parameters is implemented using a differential evolution algorithm. Using the continuous ranked probability score (CRPS) to evaluate the goodness-of-fit of our model for probabilistic forecasting, it is shown that this model often outperforms classical additive time series models, as well as homogeneous Markov switching models. We also propose a simple procedure for classifying load profiles into those with and without regime-switching behaviors.     

Parameter-efficient deep probabilistic forecasting

Probabilistic time series forecasting is crucial in many application domains, such as retail, ecommerce, finance, and biology. With the increasing availability of large volumes of data, a number of neural architectures have been proposed for this problem. In particular, Transformer-based methods achieve state-of-the-art performance on real-world benchmarks. However, these methods require a large number of parameters to be learned, which imposes high memory requirements on the computational resources for training such models. To address this problem, we introduce a novel bidirectional temporal convolutional network that requires an order of magnitude fewer parameters than a common Transformer-based approach. Our model combines two temporal convolutional networks: the first network encodes future covariates of the time series, whereas the second network encodes past observations and covariates. We jointly estimate the parameters of an output distribution via these two networks. Experiments on four real-world datasets show that our method performs on par with four state-of-the-art probabilistic forecasting methods, including a Transformer-based approach and WaveNet, on two point metrics (sMAPE and NRMSE) as well as on a set of range metrics (quantile loss percentiles) in the majority of cases. We also demonstrate that our method requires significantly fewer parameters than Transformer-based methods, which means that the model can be trained faster with significantly lower memory requirements, which as a consequence reduces the infrastructure cost for deploying these models.     

Quantile regression for the qualifying match of GEFCom2017 probabilistic load forecasting

We present a simple quantile regression-based forecasting method that was applied in the probabilistic load forecasting framework of the Global Energy Forecasting Competition 2017 (GEFCom2017). The hourly load data are log transformed and split into a long-term trend component and a remainder term. The key forecasting element is the quantile regression approach for the remainder term, which takes into account both weekly and annual seasonalities, such as their interactions. Temperature information is used only for stabilizing the forecast of the long-term trend component. Information on public holidays is ignored. However, the forecasting method still placed second in the open data track and fourth in the definite data track, which is remarkable given the simplicity of the model. The method also outperforms the Vanilla benchmark consistently.