Conference Agenda

Forecasting in industrial companies remains a challenge in times of rapidly changing environmental conditions that result in increasing uncertainty. Forecasting tasks are more and more replaced by forecast tools. Forecast results must be of highest accuracy as well as trustworthy for the user to successfully replace expert opinions as the standard to lower costs, efforts and resources.

Each forecasting method performs individually well on different types of time series. Thus, for maintaining robust and reliable forecast results, characterizing different time series for choosing the most suitable forecast method becomes a necessary step. One metric for characterizing time series is their chance of successful prediction, called predictability. For calculating predictability of a time series, several methods exist with different type of outcome, such as a continuous number or a binary. A general guideline on usage and comparison of these diverse approaches and their correlation is lacking in literature, especially in combination with forecasts validating the findings.

In our work, we combine seven different approaches for estimating the predictability of time series. We perform these tests on a synthetic benchmark dataset including random time series, linear time series and series with continuous seasonality such as sinus waves, as well as on open-source sales and stock data. For the validation of the predictability metrics with forecasting, diverse forecasting methods and evaluation metrics are used. With our results, we hope to contribute to more robust and reliable forecast tools that work reliably independent of the underlying time series.

In recent years, large scale disruptions to global supply chains, like the Covid pandemic, a ship blocking the Suez canal or sanctions against Russia, have caused production to slow down or even to stand still, causing shortages and massive losses for affected businesses. Even if only specific companies were originally affected, shortages and delays rippled along the supply network.

The established approach to deal with disruptions is to utilize safety stock and capacity to compensate for fluctuations in uncertain quantities like customer demand. This approach is tried and tested for small fluctuations. To address larger disruptions, like the above given, very high safety stock and capacity would be needed, which would lead to unnecessarily high costs.

Resilience has often been viewed as an expensive capability that drives costs. Recent studies however advocate for the development of lean resilience concepts, creating new capabilities, which enable resilience and can deal with large fluctuations, reimagining resilience from the perspectives of efficiency and value creation.

This contribution identifies gaps in current research and establishes structural deficits of approaches discussed in the literature regarding supply chains. Firstly, the need for rigorous, quantitative definitions of resilience and relevant disruptions is justified. Then, a stochastic model for aggregate production planning that includes capabilities to compensate for such large fluctuations along several dimensions is proposed. Lastly, this model is then applied to a case study pertaining to a realistic supply chain under the risk of large scale disruptions and the results of this approach are evaluated.