A Dynamic Model Averaging for the Discovery of Time-Varying Weather-Cycling Patterns

Journal article, 2021

It has been well recognized that weather variations significantly impact cycling experiences of users. However, the weather-cycling dynamic relationship over time is not well studied in the literature. In this paper, in order to bridge this gap, we propose a Dynamic Model Averaging and Dynamic Model Selection (DMA and DMS) to reveal the characteristics of time-varying responses and the associated influencing factors for young people's shared bike trips. Without loss of generality, dynamic models with unknown observational variances are also proposed. We take New York City as an instance and analyze the drifts of patterns of New York CitiBike trips under six weather factors from various aspects. The results suggest that the bike trips' responses to some weather factors fluctuate dynamically while others maintain at a relatively stable level. It is concluded that a few main influencing factors are adequate to represent the travel patterns. It is observed that dynamic models, with the strength of alleviating multicollinearity, present better forecast performance than classic models. This work can facilitate the decision makers and managers to oversee and optimise travel experience of users in real time.