Options
Essays in Causal Machine Learning
Type
doctoral thesis
Date Issued
2021-09-20
Author(s)
Abstract (De)
This doctoral thesis contains three studies in different fields of economics. The first and third chapter are in the field of operational research. The second chapter investigates a methodological question in the field of labour economics. Common to all chapters is that they deal with causal questions and that tools from the machine learning literature are used to answer these questions. The first chapter investigates a sequential contest with two players in darts, in which one of the contestants enjoys an advantage. The first-mover in those contests has potentially more, but never less moves in the match, which results in higher winning probability for the starting player. The application of causal machine learning methods shows that contestants with low performance measures and little experience benefit the most from moving first. Furthermore, contestants performing in a venue close to their hometown do not experience this advantage, unlike contestants competing in venues away from their hometown. The second chapter investigates whether methods from the machine learning toolbox for estimating the propensity score in matching estimators lead to more credible estimations. A main finding from the chapter is that the choice of the method is highly relevant. Especially for empirical applications in active labour market policy estimating the propensity score by LASSO based logit models delivers most credible results, while the usage of Random Forests may lead to a deterioration of the performance in specific situations. In the third chapter, this well performing combination of LASSO-based estimation of the propensity score in a radius-matching estimator found in the second chapter is used as estimation strategy. This work investigates schedule-related issues in the European football leagues and establishes differential winning probabilities in home matches for underdog teams depending on the schedule. The findings suggest that the current schedule favours underdog teams with fewer home matches on days that do not correspond to the usual match days.
Language
English
Keywords
Ökonometrie
Maschinelles Lernen
Arbeitsmarkt
EDIS-5106
Causal machine learning
labour market
econometrics
HSG Classification
not classified
HSG Profile Area
None
Publisher
Universität St. Gallen
Publisher place
St.Gallen
Subject(s)
Division(s)
Eprints ID
264350
File(s)