Bike Share Demand Forecasting

Forecasting Dynamics

In the realm of bike-sharing systems, understanding the dynamics between forecasting and decision-making is crucial. Daniele Gammelli emphasizes the importance of the interaction between inbound and outbound data flows, which significantly impacts decision-making processes 1. He highlights the necessity of balancing the investment in research and development with the practical utility of forecasts, noting that decision modules often exploit forecasting errors 2. This interaction requires careful management to ensure that forecasts remain robust and valuable over time.

We should always aim for what we believe will have the highest value in the end.

--- Daniele Gammelli

Kyle Polich adds that the value of information is a key consideration, as it informs decisions and guides investments in forecasting technologies 1.

   

Predictive vs. Prescriptive

The distinction between predictive and prescriptive performance in forecasting models is pivotal. Daniele Gammelli explains that while forecasting models are traditionally evaluated based on prediction accuracy, their impact on decision-making, or prescriptive performance, is often overlooked 3. He notes that the best predictive models do not always translate into the best decision-making outcomes, highlighting the need for models that integrate assumptions compatible with decision-making architectures 4. This approach ensures that forecasts are not only accurate but also actionable.

The best prediction performance is not necessarily correlated with the best decision performance downstream.

--- Daniele Gammelli

Daniele further discusses the integration of queuing theory into forecasting models to optimize inventory decisions in bike-sharing systems 4.