Dynamic predictive models for ridesourcing services in New York City using daily compositional data

Abstract

In recent years, there has been a tremendous change in the landscape of mobility offerings with the introduction with the introduction of Uber and Lyft, companies collectively known as Transportation Network Companies (TNCs), otherwise known as ridesourcing companies. Due to the nascency of these companies, there is a small but rapidly growing body of literature on the impacts these TNCs are having on traditional forms of shared ride modes, however, most of the emerging literature centers on the relationships between TNC and only one other shared mode of transportation (e.g. TNCs and Public Transit). This research attempts to contribute to the literature by examining the relationship between TNCs and multiple shared ride modes simultaneously. To this end, a joint modeling framework was used to study New York City ridership data for TNCs, taxi services, bikesharing, and the subway from January 2015 to June 2017. The goal of the research presented is two-fold: explore the dynamic relationships between TNCs and other modes of shared ridership, as well as to build a predictive model for the daily ridership usage for each modal offering and total daily ridership usage. To accomplish these tasks, we first used a compositional time series approach in which the four series are modeled as proportions of total daily demand and then, after a suitable transformation, jointly modeled them via a vector autoregression with exogenous predictors (VARX) to account for trend, a weekly seasonal structure, and exogenous predictors. The second part of our analysis involved modeling the daily total usage of the four modes using a dynamic linear model (DLM) and then using that model to draw inferences about the total ridership in NYC. Results of the models were then combined to produce medium-term forecasts for each modal ridership. Our findings corroborate those of others in investigating correlations over time between usage of TNCs and taxis in servicing consumers. In keeping with our second goal, this analysis demonstrates that our modeling framework may be useful for forecasting several competing types of shared ride modes in New York City.