Abstract:
Behavioral economics provides a framework for understanding human decision-making by inte-grating psychological insights to explain deviations from rationality. Unlike traditional economics, it acknowledges that human behavior is often constrained by cognitive limitations, influenced by emotions, and shaped by social contexts. This perspective enables the study of phenomena such as loss aversion, reference dependent preferences, and non-Bayesian judgment, where people’s intuitive assessments of probability diverge from statistical principles like Bayes’ rule.
This dissertation builds on this framework by empirically examining how immediate behavioral responses induced by deviations from rational reference points shape decision-making under uncertainty. It quantifies the average market participant’s response to new information using sports betting markets as controlled yet natural settings, distinguished by short-term exogenous terminal values independent of participant actions and isolated from broader economic risks. Additionally, the dissertation introduces a new Bayesian shrinkage prior to improve the estimation of sparse, high-dimensional models commonly associated with such contexts.
This work comprises three studies. The first examines the connection between emotions—specifically surprise and suspense—and alcohol consumption, using high-frequency transaction data on beer sales from professional soccer matches, alongside detailed in-play event data and betting odds. The findings provide robust evidence of emotional drinking, indicating that surprise leads to increased beer sales and suggesting that individuals respond more quickly to negative emotional signals than to positive ones. This study underscores the real-time link between emo-tions and consumption behavior, offering new field evidence on emotional drinking.
The second study addresses the challenges of estimating complex, high-dimensional models, where traditional methods like OLS often yield biased and inaccurate results due to overfitting, overparameterization, or violations of asymptotic assumptions. Shrinkage priors can mitigate these issues, but standard methods like the spike-and-slab and the Bayesian lasso have their limitations. To overcome these, this study introduces the Dirichlet-horseshoe prior, a novel ap-proach that combines principles from the regularized horseshoe and Dirichlet-Laplace priors to provide high accuracy and adaptive shrinkage. Simulations and empirical applications show that the Dirichlet-horseshoe can outperform standard alternatives in accuracy and precision across various sparsity levels and predictor correlations.
The final study examines how market participants and market makers respond to new infor-mation in the sports betting market. Using time-series data from multiple bookmakers over the entire betting period, the study estimates the market’s learning rate by applying GMM and Bayesian methods. The findings reveal that biased opening prices improve in accuracy as the market approaches closure. Learning does not begin immediately but intensifies over the betting period, continuing almost until closure. Learning rates differ across bookmakers and market segments, suggesting variation in the market’s ability to incorporate new information into prices.
behavioral economics