MEng Imperial College London

As part of my MEng Mathematics and Computer Science degree, my final year project involved studying the limits of Network Effects.

This work was done alongside Ali Farzanehfar and Dr. Yves-Alexandre de Montjoye in Imperial's Computational Privacy Group. Together we looked at how we could study the limits of network effects through the modelling of competition in ride-hailing platforms (e.g. Uber's growth in North America v.s. Lyft and others) using agent based graphical models.

Abstract

To give you a better idea of what we worked on, have a read of the abstract!

Hotel companies not owning a single bed, taxi companies not owning a single car, the worlds’ most diverse store not owning a single till; in the 21st century we are witnessing a rapid transformation of our way of life, greatly driven by such digital platforms. Ride hailing services in particular have been quite present in our lives, disrupting how we think about transport. As a platform business these services benefit from network effects: their value increases according to their number of users. Due to the existence of network effects, these services are believed to benefit a lot from first mover advantage. However, no one has studied whether there are limits in the network effect for ride hailing services, something crucial to competition watchdogs when deciding if they should let a service like this into the city. The current literature presents extensive analysis and modelling of various graph based methods, looking at the apparition of “small-world” or “scale-free” phenomena, as well as some structural properties. Although very useful, these models are often very abstract and rarely lend themselves to empirical falsification. Here we study the limits of network effects in ride hailing platforms. Using agent based graphical models we find the sensitivity of riders and drivers to waiting and idle time to be decisive factors in the growth of these platforms. We validate our results using real ride hailing data from the New York City Taxi and Limousine Commission dataset. We find that our model can accurately capture the growth of ride hailing platforms both in terms of market-share and population. Finally, we explore alternative "worlds" where the tension between first mover advantage and increasing waiting or idle time is readily observable. This work is a first look into how empirical findings could be used to aid data driven regulation, ultimately allowing policy makers greater insights in their decision making process.

You can learn more about our work by reading the introduction of the thesis, which you can download here.
The code for this project is open-source, check it out on GitHub!