Change point detection in financial time series in connection to purchase behaviours

Abstract

Understanding purchase behaviours of individuals is of interest when the goal is to inspire people to make more environmentally friendly choices. A company with these aspirations is Svalna AB. They have created an app that uses a carbon calculator to give an insight into greenhouse gas emissions based on financial transactions. The aim of this thesis has been to investigate purchase behaviours by comparing the underlying distributions before and after a change point has occurred. This thesis has focused on change point detection in time series using the Metropolis-Hastings algorithm. The model, which has been implemented from scratch, has been tested on well-behaved simulated time series and can accurately find a change point. It has then been used to investigate some specific cases in financial time series provided by Svalna. The results from testing on the simulated time series show a promising start and it is concluded that the overall method is a possibility to investigate the underlying distributions of financial time series.