This thesis considers the behaviour and relationships between financial and prediction markets around elections.

We begin by reviewing the literature. There are many small studies of individual elections and events, particularly of the 2016 UK European Union referendum. However, no studies that consider multiple events, nor present theories that apply in a general setting, are found. We believe this is a gap in the literature.

Chapter 1 begins with a study of the Brexit referendum. Using a flexible prior and Bayesian updating, we demonstrate a major violation of semi-strong market efficiency in both the betting and currency markets on the night following the vote. It appears that it took a full three hours for prices to reflect the information contained in the publicly available results of the referendum.

Chapter 2 presents a model linking the prices of financial and binary options in the prediction markets in the overnight session following an election. Starting from basic assumptions we find that prices in both markets should be cointegrated. Under risk neutrality the relationship is linear. However, departures from this assumption result in a non-linear cointegrating relationship. We test the theory on three recent elections. Strong support for the theory is found for two events. The linear cointegrating model fits the data from the night of the EU referendum remarkably well. However, departures from risk neutrality are needed to explain the behaviour observed on the night of the 2016 US presidential election.

Chapter 3 considers pricing relationships in the weeks and months leading up to an election. Again using economic assumptions, we derive a relationship between asset price returns and changes in the prices of betting market binary options linked to an election result. This model is extended to equities using the ubiquitous Fama–French 5 factor model. The result is a 6 factor characteristic model, where the additional factor is related to political risk. We test the model on six recent elections. Using daily data, strong support is found for the theory for four events and weak evidence for one. The remaining election does not appear to be informative for asset prices. Interesting relationships are also uncovered between firm characteristics and political sensitivity. This is achieved by exploring the political factor loadings of the different equities under study.

The main contributions of this thesis are, one, using a flexible Bayesian approach to demonstrate that, without a shadow of a doubt, any `bubble’ in opinion for remain continued well into the night of the EU referendum, and two, presenting pricing models of prediction and financial markets that apply in general settings and have strong support in the data. We also show that on nights after elections, betting markets lead financial markets on the scale of minutes to tens of minutes. This is consistent with, and an extension of, the conclusion of the existing literature that prediction markets have superior forecasting ability. Whether or not this lead–lag relationship occurs at other times prior to political events is an open research question.