Algorithm Aversion and Other Causes of Bias in Decision Behavior
Studies on Algorithm Aversion, Capital Market Forecasting, and Price Dispersion
Cumulative thesis
Date of Examination:2022-11-25
Date of issue:2022-11-28
Advisor:Prof. Dr. Markus, Spiwoks
Referee:Prof. Dr. Kilian, Bizer
Referee:Prof. Dr. Markus, Spiwoks
Referee:Prof. Dr. Holger A., Rau
Files in this item
Name:Onlineformat_Dissertation_Marco_Lorenz_20221119.pdf
Size:5.46Mb
Format:PDF
Abstract
English
The present thesis contributes to the debate on whether the neoclassical economic theory is suitable to describe economic reality. It covers the three topics of algorithm aversion, analysis of capital market forecasts, and analysis of price dispersion. The first research area in this dissertation is the rather new field of "algorithm aversion". Dietvorst et al. (2015) had shown that people tend to refrain from using highly effective automated processes (algorithms) even when their use would clearly be to their own advantage. The first paper uses an economic experiment to demonstrate that algorithm aversion can be partially reduced with increasing experience in using a prediction task and an algorithm. The second study examines the extent to which algorithm aversion is pronounced in six decision situations with different contexts. It is revealed that the aversion is particularly strong where an error can have dramatic consequences (e.g., in the detection of tumors on MRI scans). The third study examines the extent to which algorithm aversion declines when decision-makers can influence an algorithm at different points in the prediction process. The ability to adjust an algorithm's forecasts at the end of the process proves to be particularly effective. Influencing the configuration of an algorithm does not lead to a comparable reduction in algorithm aversion. Finally, studies four and five show that algorithm aversion cannot be significantly reduced either by the decoy effect or by making decisions on behalf of others. Taken as a whole, the studies show how algorithm aversion is a hindrance to the establishment of innovations such as robo-advisors and how difficult it can be to overcome it. When working with algorithms, subjects regularly deviate from the rational behavior suggested by the homo economicus model from neoclassical theory. The sixth and seventh studies examine the accuracy of interest rate and stock index forecasts. They show that developments in capital markets can mostly not be predicted with sufficient accuracy. With a few exceptions, it is therefore not to be expected that systematic excess returns can be achieved by taking the forecasts into account. The observed adherence to overly cautious and present-oriented forecasts is also at odds with the homo economicus model. Finally, the eighth study analyzes a large database of retail prices called between 2020 and 2022. The observed deviations from the neoclassical model of equilibrium price formation are so striking that here, too, it must be stated that neoclassical theory only inaccurately describes practice. Overall, the conclusion is that the popularity of neoclassical theory in research and teaching should be questioned. Subject behavior is clearly more in line with more modern approaches, such as behavioral economics, in all three topics covered in this dissertation.
Keywords: Algorithm Aversion; Behavioral Economics; Capital Markets; Equilibrium Price; Experimental Economics; Forecasting; Homo Economicus; Interest Rates; Technology Adoption