Essays in Behavioural Economics & Applied Econometrics

A long body of literature in behavioural economics has demonstrated that individuals perform poorly in understanding how past outcomes relate to future outcomes in experimental settings. This thesis studies the ability of individual decision makers to incorporate information about previous successes to predict future performance in naturalistic settings. This thesis is split into three distinct chapters. The first chapter attempts to model how past outcomes affect future outcomes in human behaviour in naturalistic settings using data on elite sports. I test if Premier League football players who score in one game are more likely to score in the next game by exploiting exogenous variation in scoring between players who took shots with the same probability of scoring. My results suggest that Premier League football players experience substantial persistence in scoring. The second chapter models how individuals react to information on past outcomes when making judgements about future outcomes under uncertainty in naturalistic settings. I study how fans react to player performances in Premier League Football, using my results from the first chapter as informative of the correct direction of reaction. I collect data on portfolio choices for over 180,000 fans participating in Fantasy Premier League over one season and link this to real-world player performances. I find that fans rationally incorporate information on players’ previous performances when making judgements about future performances. Not only do they react in the correct direction, but they also react with the correct magnitude. These results suggest that, under certain conditions, individuals can correctly react to information on past outcomes in their decision making. The third chapter studies the econometric models used to estimate the aforementioned effects. In this chapter, I describe how a statistical bias in estimating conditional probabilities from binary data, the streak selection bias, resembles a well-known bias in the dynamic panel data literature, the Nickell bias. I use Monto Carlo simulations to show that dynamic panel data estimators can be used to mitigate this bias, and I find that a bias-corrected method of moments estimator outperforms widely-used linear and non-linear dynamic panel data estimators in estimating the effect of interest. I use this estimator to reproduce a finding in the literature, and suggest cases in which this estimator may find use in practice.<p></p>

Author dc:creator

• Scott Dickinson (21042194)

• Information and Beliefs
• Choice Under Uncertainty
• Dynamic Panel Data

Statement dc:rights

• All rights reserved
• Open Access after 2029-09-30

Identifier

10779/exe.32542155.v1

OAI identifier oai:identifier

oai:figshare.com:article/32542155