University of Southampton
Machine Learning and Econometrics Models for Behavioural Decision-Making
Abstract
dc:description.abstractMachine Learning (ML) and Econometrics models are a powerful tool for developing and testing theories by way of prediction, causal explanation and description. In many businesses, the priority and focus are; do I predict, or do I explain? Nearly all decisions are based on predictions and causal explanation, whether more intuitive or deliberative. This thesis, which is divided into three papers, explores the use of prediction methods and causal explanation, with applications to financial markets and marketing/e-commerce. <br/><br/>In paper one, we compared the accuracy of deep and shallow architectures by predicting thirty-four different stock indices across different time horizons (daily, hourly, minute and tick level) using financial market data. We contribute to the ML literature by exploring the degree to which is possible to predict stock price indices across different time horizon and markets. <br/><br/>Paper two explore the use of behavioural data from a large retail organisation to <br/>understand cross device browsing behaviour. These allows us to fill an important gap that exists in marketing/e-commerce literature, about which platform offers higher conversion rates and how online consumer browsing and buying behaviour differs among these platforms. Despite conversion rate across devices, a large proportion of consumers leave items in their shopping cart without completing a purchase in a session. This is referred to as cart abandonment. Industry report shows the rate of cart abandonment across all sectors of 75.6%.<br/><br/>In paper 3, we developed a unified framework using a recursive bivariate probit (RBP) model to explain the differences in online shopping cart abandonment across mobile and nonmobile devices. To our knowledge, extant research has not examined online shopping cart abandonment across mobile and non-mobile devices with field data or e-commerce click stream data. Our framework used features such as when shoppers have high basket values, browsing in the evening, if reading reviews on mobile vs non-mobile channel and numbers of attempted credit card failures on mobile vs non-mobile channel to understand device differences.<br/><br/>In summary, this thesis offers insights into decision-making using prediction methods at the algorithmic and individual level.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D.
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Southampton
- Year dc:date.issued
- 2021
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Orimoloye, Olanrewaju
- Advisors dc:contributor.advisor
-
- Ma, Tiejun
- Johnson, Johnnie
- Sung, Ming-Chien
Rights
- Language dc:language
- English, eng