UCLA

As the ride-hailing economy has been rapidly growing, case studies of Uber’s forecasting and pricing frameworks have received much attention among researchers and investigative journalists. This paper 1) attempts to forecast Uber’s fulfilled pickups via predictive modelling, and 2) conducts a retrospective study to examine the role of socio-economic factors and surge pricing on riders’ waiting time along with their implications. By seeking to answer questions regarding these two topics, this paper hopes to shed light on the demand-supply dynamics of Uber service, and provides empirical analysis and potential suggestions for future improvement. The scope of this paper focuses on New York City and Washington D.C., where public data is available. This paper evaluates the performance under various predictive statistical learning models based on the lowest error rate (RMSE being the major criteria), while the explanatory modelling section adopts more traditional statistical tools, as well as a Bayesian network. The results suggest that LASSO has outperformed other models used in this thesis for the demand forecast. Moreover, there is a lack of evidence that surge pricing has a consistent effect in reducing riders’ average waiting time. Furthermore, various socio-economic factors are found to have significant associations with waiting time, and in particular, districts with higher ethnic minority rates are associated with longer average waiting time, which calls for further examination on the dynamics of Uber’s service delivery.