Retail forecasting: Research and practice

This paper reviews the research literature on forecasting retail demand. We begin by introducing the forecasting problems that retailers face, from the strategic to the operational, as sales are aggregated over products to stores and to the company overall. Aggregated forecasting supports strategic decisions on location. Product-level forecasts usually relate to operational decisions at the store level. The factors that influence demand, and in particular promotional information, add considerable complexity, so that forecasters potentially face the dimensionality problem of too many variables and too little data. The paper goes on to evaluate evidence on comparative forecasting accuracy. Although causal models outperform simple benchmarks, adequate evidence on machine learning methods has not yet accumulated. Methods for forecasting new products are examined separately, with little evidence being found on the effectiveness of the various approaches. The paper concludes by describing company forecasting practices, offering conclusions as to both research gaps and barriers to improved practice.     

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.     

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.     

Using weather ensemble predictions in electricity demand forecasting

Weather forecasts are an important input to many electricity demand forecasting models. This study investigates the use of weather ensemble predictions in electricity demand forecasting for lead times from 1 to 10 days ahead. A weather ensemble prediction consists of 51 scenarios for a weather variable. We use these scenarios to produce 51 scenarios for the weather-related component of electricity demand. The results show that the average of the demand scenarios is a more accurate demand forecast than that produced using traditional weather forecasts. We use the distribution of the demand scenarios to estimate the demand forecast uncertainty. This compares favourably with estimates produced using univariate volatility forecasting methods.     

Predicting/hypothesizing the findings of the M5 competition

The scientific method consists of making hypotheses or predictions and then carrying out experiments to test them once the actual results have become available, in order to learn from both successes and mistakes. This approach was followed in the M4 competition with positive results and has been repeated in the M5, with its organizers submitting their ten predictions/hypotheses about its expected results five days before its launch. The present paper presents these predictions/hypotheses and evaluates their realization according to the actual findings of the competition. The results indicate that well-established practices, like combining forecasts, exploiting explanatory variables, and capturing seasonality and special days, remain critical for enhancing forecasting performance, re-confirming also that relatively new approaches, like cross-learning algorithms and machine learning methods, display great potential. Yet, we show that simple, local statistical methods may still be competitive for forecasting high granularity data and estimating the tails of the uncertainty distribution, thus motivating future research in the field of retail sales forecasting.     

Forecasting third-party mobile payments with implications for customer flow prediction

Forecasting customer flow is key for retailers in making daily operational decisions, but small retailers often lack the resources to obtain such forecasts. Rather than forecasting stores’ total customer flows, this research utilizes emerging third-party mobile payment data to provide participating stores with a value-added service by forecasting their share of daily customer flows. These customer transactions using mobile payments can then be utilized further to derive retailers’ total customer flows indirectly, thereby overcoming the constraints that small retailers face. We propose a third-party mobile-payment-platform centered daily mobile payments forecasting solution based on an extension of the newly-developed Gradient Boosting Regression Tree (GBRT) method which can generate multi-step forecasts for many stores concurrently. Using empirical forecasting experiments with thousands of time series, we show that GBRT, together with a strategy for multi-period-ahead forecasting, provides more accurate forecasts than established benchmarks. Pooling data from the platform across stores leads to benefits relative to analyzing the data individually, thus demonstrating the value of this machine learning application.     

News will tell: Forecasting foreign exchange rates based on news story events in the economy calendar

The paper proposes a novel approach to predict intraday directional-movements of currency-pairs in the foreign exchange market based on news story events in the economy calendar. Prior work on using textual data for forecasting foreign exchange market developments does not consider economy calendar events. We consider a rich set of text analytics methods to extract information from news story events and propose a novel sentiment dictionary for the foreign exchange market. The paper shows how news events and corresponding news stories provide valuable information to increase forecast accuracy and inform trading decisions. More specifically, using textual data together with technical indicators as inputs to different machine learning models reveals that the accuracy of market predictions shortly after the release of news is substantially higher than in other periods, which suggests the feasibility of news-based trading. Furthermore, empirical results identify a combination of a gradient boosting algorithm, our new sentiment dictionary, and text-features based-on term frequency weighting to offer the most accurate forecasts. These findings are valuable for traders, risk managers and other consumers of foreign exchange market forecasts and offer guidance how to design accurate prediction systems.     

Applications of machine learning for corporate bond yield spread forecasting

This article considers nine different predictive techniques, including state-of-the-art machine learning methods for forecasting corporate bond yield spreads with other input variables. We examine each method’s out-of-sample forecasting performance using two different forecast horizons: (1) the in-sample dataset over 2003–2007 is used for one-year-ahead and two-year-ahead forecasts of non-callable corporate bond yield spreads; and (2) the in-sample dataset over 2003–2008 is considered to forecast the yield spreads in 2009. Evaluations of forecasting accuracy have shown that neural network forecasts are superior to the other methods considered here in both the short and longer horizon. Furthermore, we visualize the determinants of yield spreads and find that a firm’s equity volatility is a critical factor in yield spreads.     

The tourism forecasting competition

We evaluate the performances of various methods for forecasting tourism data. The data used include 366 monthly series, 427 quarterly series and 518 annual series, all supplied to us by either tourism bodies or academics who had used them in previous tourism forecasting studies. The forecasting methods implemented in the competition are univariate and multivariate time series approaches, and econometric models. This forecasting competition differs from previous competitions in several ways: (i) we concentrate on tourism data only; (ii) we include approaches with explanatory variables; (iii) we evaluate the forecast interval coverage as well as the point forecast accuracy; (iv) we observe the effect of temporal aggregation on the forecasting accuracy; and (v) we consider the mean absolute scaled error as an alternative forecasting accuracy measure. We find that pure time series approaches provide more accurate forecasts for tourism data than models with explanatory variables. For seasonal data we implement three fully automated pure time series algorithms that generate accurate point forecasts, and two of these also produce forecast coverage probabilities which are satisfactorily close to the nominal rates. For annual data we find that Naïve forecasts are hard to beat.     

Automatic Interpretable Retail forecasting with promotional scenarios

Budgeting and planning processes require medium-term sales forecasts with marketing scenarios. The complexity in modern retailing necessitates consistent, automatic forecasting and insight generation. Remedies to the high dimensionality problem have drawbacks; black box machine learning methods require voluminous data and lack insights, while regularization may bias causal estimates in interpretable models.The proposed FAIR (Fully Automatic Interpretable Retail forecasting) method supports the retail planning process with multi-step-ahead category-store level forecasts, scenario evaluations, and insights. It considers category-store specific seasonality, focal- and cross-category marketing, and adaptive base sales while dealing with regularization-induced confounding.We show, with three chains from the IRI dataset involving 30 categories, that regularization-induced confounding decreases forecast accuracy. By including focal- and cross-category marketing, as well as random disturbances, forecast accuracy is increased. FAIR is more accurate than the black box machine learning method Boosted Trees and other benchmarks while also providing insights that are in line with the marketing literature.     

LoMEF: A framework to produce local explanations for global model time series forecasts

Global forecasting models (GFMs) that are trained across a set of multiple time series have shown superior results in many forecasting competitions and real-world applications compared with univariate forecasting approaches. One aspect of the popularity of statistical forecasting models such as ETS and ARIMA is their relative simplicity and interpretability (in terms of relevant lags, trend, seasonality, and other attributes), while GFMs typically lack interpretability, especially relating to particular time series. This reduces the trust and confidence of stakeholders when making decisions based on the forecasts without being able to understand the predictions. To mitigate this problem, we propose a novel local model-agnostic interpretability approach to explain the forecasts from GFMs. We train simpler univariate surrogate models that are considered interpretable (e.g., ETS) on the predictions of the GFM on samples within a neighbourhood that we obtain through bootstrapping, or straightforwardly as the one-step-ahead global black-box model forecasts of the time series which needs to be explained. After, we evaluate the explanations for the forecasts of the global models in both qualitative and quantitative aspects such as accuracy, fidelity, stability, and comprehensibility, and are able to show the benefits of our approach.     

An accurate and fully-automated ensemble model for weekly time series forecasting

Many businesses and industries require accurate forecasts for weekly time series nowadays. However, the forecasting literature does not currently provide easy-to-use, automatic, reproducible and accurate approaches dedicated to this task. We propose a forecasting method in this domain to fill this gap, leveraging state-of-the-art forecasting techniques, such as forecast combination, meta-learning, and global modelling. We consider different meta-learning architectures, algorithms, and base model pools. Based on all considered model variants, we propose to use a stacking approach with lasso regression which optimally combines the forecasts of four base models: a global Recurrent Neural Network (RNN) model, Theta, Trigonometric Box–Cox ARMA Trend Seasonal (TBATS), and Dynamic Harmonic Regression ARIMA (DHR-ARIMA), as it shows the overall best performance across seven experimental weekly datasets on four evaluation metrics. Our proposed method also consistently outperforms a set of benchmarks and state-of-the-art weekly forecasting models by a considerable margin with statistical significance. Our method can produce the most accurate forecasts, in terms of mean sMAPE, for the M4 weekly dataset among all benchmarks and all original competition participants.     

Forecasting small area populations with long short-term memory networks

Local and state governments depend on small area population forecasts to make important decisions concerning the development of local infrastructure and services. Despite their importance, current methods often produce highly inaccurate forecasts. Recent years have witnessed promising developments in time series forecasting using Machine Learning across a wide range of social and economic variables. However, limited work has been undertaken to investigate the potential application of Machine Learning methods in demography, particularly for small area population forecasting. In this paper we describe the development of two Long-Short Term Memory network architectures for small area populations. We employ the Keras Tuner to select layer unit numbers, vary the window width of input data, and apply a double training and validation regime which supports work with short time series and prioritises later sequence values for forecasts. These methods are transferable and can be applied to other data sets. Retrospective small area population forecasts for Australia were created for the periods 2006–16 and 2011–16. Model performance was evaluated against actual data and two benchmark methods (LIN/EXP and CSP-VSG). We also evaluated the impact of constraining small area population forecasts to an independent national forecast. Forecast accuracy was influenced by jump-off year, constraining, area size, and remoteness. The LIN/EXP model was the best performing method for the 2011-based forecasts whilst deep learning methods performed best for the 2006-based forecasts, including significant improvements in the accuracy of 10 year forecasts. However, benchmark methods were consistently more accurate for more remote areas and for those with populations <5000.     

Machine learning in M4: What makes a good unstructured model?

The Makridakis Competitions seek to identify the most accurate forecasting methods for different types of predictions. The M4 competition was the first in which a model of the type commonly described as “machine learning” has outperformed the more traditional statistical approaches, winning the competition. However, many approaches that were self-labeled as “machine learning” failed to produce accurate results, which generated discussion about the respective benefits and drawbacks of “statistical” and “machine learning” approaches. Both terms have remained ill-defined in the context of forecasting. This paper introduces the terms “structured” and “unstructured” models to better define what is intended by the use of the terms “statistical” and “machine learning” in the context of forecasting based on the model’s data generating process. The mechanisms that underlie specific challenges to unstructured modeling are examined in the context of forecasting, along with common solutions. Finally, the innovations in the winning model that allowed it to overcome these challenges and produce highly accurate results are highlighted.     

Real-time inflation forecasting using non-linear dimension reduction techniques

In this paper, we assess whether using non-linear dimension reduction techniques pays off for forecasting inflation in real-time. Several recent methods from the machine learning literature are adopted to map a large dimensional dataset into a lower-dimensional set of latent factors. We model the relationship between inflation and the latent factors using constant and time-varying parameter (TVP) regressions with shrinkage priors. Our models are then used to forecast monthly US inflation in real-time. The results suggest that sophisticated dimension reduction methods yield inflation forecasts that are highly competitive with linear approaches based on principal components. Among the techniques considered, the Autoencoder and squared principal components yield factors that have high predictive power for one-month- and one-quarter-ahead inflation. Zooming into model performance over time reveals that controlling for non-linear relations in the data is of particular importance during recessionary episodes of the business cycle or the current COVID-19 pandemic.     

Machine learning advances for time series forecasting

In this paper, we survey the most recent advances in supervised machine learning (ML) and high-dimensional models for time-series forecasting. We consider both linear and nonlinear alternatives. Among the linear methods, we pay special attention to penalized regressions and ensemble of models. The nonlinear methods considered in the paper include shallow and deep neural networks, in their feedforward and recurrent versions, and tree-based methods, such as random forests and boosted trees. We also consider ensemble and hybrid models by combining ingredients from different alternatives. Tests for superior predictive ability are briefly reviewed. Finally, we discuss application of ML in economics and finance and provide an illustration with high-frequency financial data.     

Considerations of a retail forecasting practitioner

Forecasts can be used in an extraordinarily diverse range of ways across many domains in which forecasting practitioners work continuously towards improving their forecasts. Each of these domains may require the analysis of different kinds of inputs and special considerations. Even within a given domain, such as retail, there may be many similar use cases of the same kind of forecast, which can lead to practitioners making different decisions. This paper discusses several of the important decision points that practitioners must work through and uses item-level sales forecasting in the retail domain as leveraged by pricing and inventory management as examples of the different paths that may be taken. It considers how each use can lead to a different forecasting objective, and a corresponding focus on different error metrics. In addition, there are several tradeoffs in the forecasting methods that are used to meet each of the objectives best, including the kinds of models used, the running time speed, and forecast accuracy requirements.     

Comparison of intraday probabilistic forecasting of solar irradiance using only endogenous data

Accurate solar forecasts are necessary to improve the integration of solar renewables into the energy grid. In recent years, numerous methods have been developed for predicting the solar irradiance or the output of solar renewables. By definition, a forecast is uncertain. Thus, the models developed predict the mean and the associated uncertainty. Comparisons are therefore necessary and useful for assessing the skill and accuracy of these new methods in the field of solar energy.The aim of this paper is to present a comparison of various models that provide probabilistic forecasts of the solar irradiance within a very strict framework. Indeed, we consider focusing on intraday forecasts, with lead times ranging from 1 to 6 h. The models selected use only endogenous inputs for generating the forecasts. In other words, the only inputs of the models are the past solar irradiance data. In this context, the most common way of generating the forecasts is to combine point forecasting methods with probabilistic approaches in order to provide prediction intervals for the solar irradiance forecasts. For this task, we selected from the literature three point forecasting models (recursive autoregressive and moving average (ARMA), coupled autoregressive and dynamical system (CARDS), and neural network (NN)), and seven methods for assessing the distribution of their error (linear model in quantile regression (LMQR), weighted quantile regression (WQR), quantile regression neural network (QRNN), recursive generalized autoregressive conditional heteroskedasticity (GARCHrls), sieve bootstrap (SB), quantile regression forest (QRF), and gradient boosting decision trees (GBDT)), leading to a comparison of 20 combinations of models.None of the model combinations clearly outperform the others; nevertheless, some trends emerge from the comparison. First, the use of the clear sky index ensures the accuracy of the forecasts. This derived parameter permits time series to be deseasonalized with missing data, and is also a good explanatory variable of the distribution of the forecasting errors. Second, regardless of the point forecasting method used, linear models in quantile regression, weighted quantile regression and gradient boosting decision trees are able to forecast the prediction intervals accurately.     

Forecasting and turning point predictions in a Bayesian panel VAR model

We provide methods for forecasting variables and predicting turning points in panel Bayesian VARs. We specify a flexible model, which accounts for both interdependencies in the cross section and time variations in the parameters. Posterior distributions for the parameters are obtained for hierarchical and for Minnesota-type priors. Formulas for multistep, multiunit point and average forecasts are provided. An application to the problem of forecasting the growth rate of output and of predicting turning points in the G-7 illustrates the approach. A comparison with alternative forecasting methods is also provided.     

FFORMPP: Feature-based forecast model performance prediction

This paper introduces a novel meta-learning algorithm for time series forecast model performance prediction. We model the forecast error as a function of time series features calculated from historical time series with an efficient Bayesian multivariate surface regression approach. The minimum predicted forecast error is then used to identify an individual model or a combination of models to produce the final forecasts. It is well known that the performance of most meta-learning models depends on the representativeness of the reference dataset used for training. In such circumstances, we augment the reference dataset with a feature-based time series simulation approach, namely GRATIS, to generate a rich and representative time series collection. The proposed framework is tested using the M4 competition data and is compared against commonly used forecasting approaches. Our approach provides comparable performance to other model selection and combination approaches but at a lower computational cost and a higher degree of interpretability, which is important for supporting decisions. We also provide useful insights regarding which forecasting models are expected to work better for particular types of time series, the intrinsic mechanisms of the meta-learners, and how the forecasting performance is affected by various factors.