ISF 2023 Workshop | Business Forecasting: Techniques, Application and Best Practices

Part I: Introduction to Forecasting

A broad overview of business forecasting and its various uses within the organization. Topics include approaches to forecasting, features of data, the role of judgment, and resources for forecasters.

Part II: Univariate Forecasting

Exponential Smoothing
A survey of exponential smoothing techniques with particular emphasis on the Holt-Winters family of models, Croston’s intermittent demand model and a model designed to forecast items that exhibit significant volume only at certain times of the year. Topics include the pros and cons of using these models, when they are best used, how they work, identifying model components, parameter optimization and model diagnosis.

Box-Jenkins (ARIMA) Models
An exploration into the use of ARIMA models for business forecasting. Topics include the advantages/disadvantages of using these models, how and when they should be applied, automatic identification procedures, and model diagnostics.

Part III: Multivariate Forecasting

A discussion of the benefits and requirements for using multivariate forecasting methods followed by a closer look at event-index models, machine learning approaches, and dynamic regression models.

Event-index Models
Event-index models extend the functionality of exponential smoothing models by providing adjustments for promotions, stock outs and other events that move around the calendar. This unit addresses how these models work, how and when they should be used, and how to customize their design to best suit your needs.

Forecasting with Machine Learning
An overview of the basics and benefits of forecasting with machine learning (ML). Topics include the basics of machine learning powered forecasting, when ML is likely to improve your forecasts, and the steps involved in generating ML forecasts. A particular emphasis will be placed on extreme gradient boosted trees, an ML approach that performed well in the M5 forecasting competition.

Dynamic Regression
A detailed look into the ins and outs of regression forecasting. Topics include when regression models are best applied, how to build a regression model, ordinary least squares, leading indicators, lagged variables, Cochrane-Orcutt models, hypothesis testing and the use of “dummy” variables.

Part IV: Post Workshop Video Access

In addition to the live workshop, attendees will have two weeks of access to an on-demand video library of the topics that will not be presented live. These include:

Components of Data
An in-depth look at the different components found in time series data including trends, seasonal patterns, business cycles, trading-day variations, interventions (events) and noise. Discussion includes the forms the components can take, spotting local vs. global components, interpretation of business cycle indicators and the use of decomposition routines.

Forecasting Accuracy and Evaluation
A detailed look at evaluating the accuracy of forecasting methods. Topics include the distinction between within-sample and out-of-sample errors, a survey of error measurement statistics, a summary of findings from forecasting competitions and an explanation of how to use both real-time tracking reports and simulations as predictors of model performance.

Identifying Problems in Your Forecasting Process
Approaches for focusing on critical items when forecasting large volumes of data. Topics include evaluating and forecasting SKU data, filtering and ABC (Pareto) classification, outlier detection and correction, exception reporting and measuring accuracy across multiple time series.

Multiple-Level Forecasting
This session explores hierarchical forecasting techniques. Topics include discussion of the need for forecasting at various levels, product vs. geographical hierarchies, reconciliation strategies, top-down vs. bottom-up approaches, the use of proportional allocation and adjustment for seasonality.

New Product Forecasting
This session explores various approaches for forecasting new products. Topics include the pros and cons of different methods based on a product’s classification, and a review of popular methods including item supersession, forecasting by analogy and the Bass diffusion model.