Algorithmic Decision-Making, Economic Behavior and Predictability in Financial Markets

Abstract

Digital transformation is producing a growing number of technological innovations that have an impact on our daily lives. In a variety of areas, economic agents increasingly have the opportunity to interact with algorithms, as shown, for example, by the offering of robo-advisors, and thus also to influence events on financial markets. This thesis aims to examine the behavior of economic agents when interacting with algorithms and their willingness to use them in order to contribute to a better understanding of Algorithm Aversion. Algorithm Aversion describes the negative attitude towards the use of algorithms that economic agents often exhibit once they realize that algorithms are superior but not error-free. The first part of this thesis consists of five experimental studies in this regard. The first contribution shows that Algorithm Aversion in repeated tasks can be partially reduced by increasing experience over time. The second contribution addresses the scope of a decision and shows that the use of algorithms is often rejected in situations where the consequences of an error are serious, even though their use has a higher chance of success. The third contribution shows that possible user interventions in the prediction generation process reduce Algorithm Aversion more reliably if they are granted on the prediction result (output of the algorithm) instead of on the configuration (input of an algorithm). The fourth contribution examines the impact of proxy decisions on Algorithm Aversion. However, making decisions for third parties does not reduce the extent of Algorithm Aversion. The fifth contribution shows that the decision behavior to use an algorithm varies with the prior adoption rate of other economic agents, and that a prior high adoption rate leads to more frequent use of an algorithm than a prior low adoption rate. Overall, Algorithm Aversion proves to be highly robust and can contribute to suboptimal decisions. Overcoming Algorithm Aversion is essential to exploit the great potential that technological innovation brings to forecasting. The second part of this thesis consists of two more papers that contribute to the literature on the quality of capital market forecasts. While the sixth contribution examines the quality of interest rate forecasts in the Latin American region, the seventh contribution focuses on stock market forecasts for three major indices. Overall, the capital market forecasts examined are in most cases inadequate. While forecasts in the Latin American region largely reflect current rather than future interest rate developments, stock index forecasts show that most stock market analysts underestimate the variability of reality and tend toward conservatism. Therefore, it is crucial to improve forecasting models and to react more flexibly to new developments. vii Zusammenfassung Die digitale Transformation bringt immer mehr technische Innovationen hervor, die Auswirkungen auf unser tägliches Leben haben. In einer Vielzahl von Bereichen haben Wirtschaftsakteure zunehmend die Möglichke...