On beliefs and learning in investment decisions

Whether we enjoy the fruits of a good decision or experience the fallout of a bad one, we are always haunted by our past to some extent. Whatever the outcome may be, it is crucial to learn from this past, especially if a new, maybe similar, decision awaits. The question arises which aspects can be used to improve the expected outcome of the new decision and which aspects are better disregarded. This thesis consists of three manuscripts in which I investigate how this question is answered in the domain of investment decisions. In doing so I also explore three potential remedies to averse effects of clinging to prior decisions: Improved information, increased mental distance and decreased personal involvement.
The first manuscript looks specifically at how previous gains or losses influence the way in which a decision maker learns from a new signal. I show that this investment position (i.e., whether the investment has created a gain or a loss so far) interacts with the favorability of the new information (i.e., whether it is good or bad news). This interaction can lead to beliefs that may underlie well known and profit harming investment behaviors such as the Disposition Effect. I develop an extended Reinforcement Learning model that captures this interaction in the learning rate and show that it represents the data of an investment experiment better than competing models. In a second part of the experiment I also show that this interaction effect can only be overcome if participants are provided with very clear information such as revealing the true probability of the next price move.
The second manuscript builds upon the findings of the first. It presents a different investment experiment in which I aim to improve participants' belief updating by increasing the perceived mental distance from their initial investment decision as well as the development of their investment. In the first of two treatment conditions participants were blocked from changing their investment for a number of rounds. In the second treatment condition participants could neither change their investment nor track its development for the same number of periods. I show that in the treatment conditions participants' beliefs indeed approach those of a Bayesian updater calculating the objective probabilities. However, only the condition in which both the investment and information was blocked was sufficient to render this improvement significant.
Finally, the third manuscript continues the theme of involvement. It uses a sample of professional decision makers to replicate a study aimed at investigating the sunk cost fallacy. In the experimental task, participants could choose one of six lotteries. These lotteries were either framed as an investment into a startup or as the hiring of a consultant, that would then invest on the participants' behalf. Participants were generally less likely to change their chosen investment option after a bad outcome if their previous decision was framed as their own investment (i.e. they were directly involved) rather than as hiring a consultant (and thereby having "someone to blame"). Additionally I find that this effect was only significant for younger participants, which could indicate that these negative effects of involvement may fade with increasing experience.