The ISF offers unique, tailored workshops for symposium registrants. These workshops offer the opportunity to participate in an in-depth look at a specific forecasting theme. Workshops take place on Sunday, June 28. When registering for the symposium, you will be given the option to select workshop(s). The workshop fees are US$75 (1/2 day) or $150 (full day).

If you have already registered and would like to add a workshop, go to the registration page. Once here, simply enter your personal information, and under the registration options, select ‘Already registered? Select this option to add workshop‘

Workshop 1: Forecasting with Temporal Hierarchies
Instructor: Nikolaos Kourentzes, Head of Research at Indicio Technologies
9am – 12pm

Forecasting with Temporal Hierarchies (THieF) is a relatively new method in time series forecasting. Initially we introduced it as a tool to improve forecast accuracy, irrespective of the selected forecasting models. Nonetheless, THieF goes beyond due to its connections (i) with hierarchical forecasting from a technical perspective, and (ii) with decision making from an organizational perspective, as it connects short- and long-term projections and plans.

Workshop 2: How to make sure that your forecasts make sense? Workshop on forecasts evaluation.
Instructor: Ivan Svetunkov, Lancaster University
1pm – 4pm

In this workshop, we will discuss how to correctly evaluate forecasts to make sure that your favorite model still makes sense. We will start from the discussion of forecast errors and their properties, then the evaluation schemes, after which we will move to the point forecasts evaluation, finishing the workshop with the predictive distribution evaluation. The workshop will rely on the academic papers written in the area of forecast evaluation and will discuss the state of the art in the area. We currently do not plan to use any software in the workshop, but we will demonstrate some principles using Excel, R and Python and will provide a list of packages implementing the discussed approaches.

Workshop 3: Forecasting to meet demand
Instructor: Stephan Kolassa, SAP Switzerland AG
9am – 12pm

Demand is influenced by various drivers, from the “standard” ones treated in every forecasting course and textbook like seasonality, trend etc. to causal factors we can influence, like prices or promotions, to factors we cannot influence, like the weather or a competitors’ marketing activities. We will discuss forecasting as one ingredient into other processes, data and data quality (with a particular emphasis on causal drivers), the forecasting process itself and forecast quality measurement. We will conclude with a summary of lessons we have learned (the hard way, partly). We will not dig into specific forecasting models or discuss specific software. Instead, we will focus on the larger picture and work in a model-agnostic way so you can apply what you learn whether your model of choice is ordinary least squares, a transformer, or boosting.

Workshop 4: Forecast combinations: decomposition, ensembles, and reconciliation
Instructor: Mitchell O’Hara-Wild, Monash University
1pm – 4pm

Combining forecasts from multiple models, time series, or components is a proven strategy for enhancing the accuracy and reliability of time series forecasts. This workshop provides an introduction to three forecast combination techniques: decomposition forecasting, ensembling, and forecast reconciliation. Decomposition forecasting combines independent forecasts of each component (often trend, seasonal, and remainder) enabling simpler models to quickly produce forecasts at scale. Ensemble forecasting leverages the strengths of multiple models to create a more robust and accurate forecast, while forecast reconciliation utilizes information across structurally related time series to improve accuracy with coherent forecasting. Learn the theory and practice of combining forecasts with hands-on exercises in R with packages from the tidy time series forecasting ecosystem (featuring tsibble, fable, feasts, and distributional).

Workshop 5: Comprehensive Workshop on Linear State Space Modeling
Instructors: Rajesh Selukar, SAS Institute
9am – 4pm

This workshop offers a comprehensive and accessible introduction to Linear State Space Models (SSMs), a powerful and interpretable framework for analyzing sequential data such as univariate and multivariate time series, as well as longitudinal and panel data. Designed for researchers and practitioners working with time series data, the session aims to bridge theoretical foundations with practical applications using real-world examples.

Workshop 6: AI in Model Building and Forecasting ~ Which part of AI hype is really of interest for Model Building and Forecasting in Econometrics
Instructors: Dr. Hans Georg Zimmermann, Chief Scientist at Fraunhofer Society
9am – 4pm

The current AI hype is impressive, but at ISF our focus is not on LLMs but on time-series forecasting. Examining the underlying mathematics of AI, we must ask: which developments are genuinely useful for data analysis, particularly for time series? This abstract highlights two key topics. [click above link for a full description]

Workshop 7: Business Forecasting: Techniques, Application and Best Practices
Instructors: Eric Stellwagen, Sarah Darin and Franklin Rea, Business Forecast Systems
9am – 4pm

This workshop provides an overview of widely used business forecasting methods, explaining how they work conceptually, their strengths and limitations, and best practices for applying them in a business environment. Numerous real-life examples from a range of industries will be presented. You will leave the workshop with a working knowledge of quantitative and qualitative forecasting methods, enabling you to improve your forecast process and your forecast accuracy. [click above link for a full description]

Workshop 8: Deep Learning & Foundation Models for Forecasting
Instructors: Tim Januschowski, Databricks; Kashif Rasul, Caner Türkmen, KeyStone AI
9am – 12pm

In this in-person workshop, we aim to cover deep forecasting methods from the ground up, starting from the very basics of deep learning to well-established deep forecasting model such as DeepAR (Salinas et al., 2019) to more recent forecasting models, including foundation models for which we’ll review selected models.

The workshop will be in-person, with a mix of theoretical lectures and practical sessions. In the lectures, we will focus on the fundamentals of deep learning such as the various architecture types (e.g. feed-forward, convolutional, recurrent neural networks and transformers) and the most important breakthroughs that established the strength of neural networks. Then, we will see how deep learning can be applied to forecasting by reviewing several state-of-the-art neural forecasting models (e.g., WaveNet (Van Den Oord et al., 2016), DeepAR (Salinas et al., 2020), NBEATS (Oreshkin et al., 2019) and the sequence-to-sequence model family [7, 10]). We will introduce AutoGluon-Timeseries, (Shchur et al., 2023) an open-source toolkit for easily training highly accurate probabilistic forecasting models, which automatically combines and tunes both statistical forecasting and deep learning models from frameworks such as GluonTS (Alexandrov et al., 2020) and in particular foundation models like Chronos and Llag-lama which we will also review in some depth.

To complement the lectures, we will offer practical sessions for the workshop participants where we will rely on AutoGluon.